Monash 80129 PDF

A Study of the Failure of
Buried Reticulation Pipes in

Reactive Soils
Thesis in Fulfilment of the Requirements for the Degree of
Doctor of Philosophy
By
Scott John Francis Gould

BEng (Hons), BBusAdmin
Department of Civil Engineering

Monash University
Clayton, Australia
August 2011
Dedication
To my Wife Jane, my Family and my Friends.
“If we knew what it was we were doing,

it would not be called research,
would it?”
Albert Einstein
iii
Copyright Notices
Notice 1
Under the Copyright Act 1968, this thesis must be used only under the normal
conditions of scholarly fair dealing. In particular no results or conclusions
should be extracted from it, nor should it be copied or closely paraphrased in
whole or in part without the written consent of the author. Proper written
acknowledgement should be made for any assistance obtained from this thesis.
Notice 2
I certify that I have made all reasonable efforts to secure copyright permissions
for third-party content included in this thesis and have not knowingly added
copyright content to my work without the owner’s permission.
v
Declaration
I hereby declare that this thesis contains no material which has been accepted
for the award of any other degree or diploma at any university or equivalent
institution and that, to the best of my knowledge and belief, this thesis con-
tains no material previously published or written by another person, except
where due reference is made in the text of the thesis. Where sections of this
thesis include the results of joint research or scholarly publication clear ac-
knowledgement of the relative contributions of the respective authors is made.
Scott John Francis Gould

Department of Civil Engineering
Monash University
Clayton, Australia
vii
Executive Summary
The failure of buried reticulation pipes has been reported to peak during winter
months in a number of countries, including Canada and the United Kingdom,
prompting much research. In Australia, the number of pipe failures has been
reported to peak during summer. However, this peak does not occur in a con-
sistent manner, varying in magnitude to such a degree as to be strongly evident
in some years and scarcely observable in others. No quantitative explanation
of the cause of such variation has, as yet been published. Consequently, little is
known about the specific causes of seasonal failures in the Australian context.
This thesis investigates the causes of the seasonal variations in the failure
of Australian water reticulation pipes. An exploratory statistical analysis of
historical data was undertaken and used as the basis to develop the hypothesis
that the seasonal variation in pipe failure numbers in Melbourne, Australia oc-
curs as the result of soil shrinkage. A detailed field study was then undertaken
on an in-service pipe and its surrounding environment to test this hypothesis.
Analysis of the data collected during the field study supported this hypothesis.
A model to represent the mechanism by which soil shrinkage results in
the development of pipe flexural stress is also presented. This model enables
the knowledge gained from the field study to be generalised and applied else-
where. The model uses a novel constitutive surface to determine soil stiffness
and hydric expansion coefficient, and a numerical sub-model to determine the
equilibrium state of the pipe-soil system. Validation of the model against data
collected during the field study showed good agreement.
This thesis has improved the understanding of the causes of the failure of
buried water reticulation pipe, specifically focusing on the interrelated factors
causing the seasonal variation of buried water reticulation pipe failures. This
improved understanding will assist asset managers by enabling them to identify
assets at high risk of failure due to environmental and climatic conditions.
ix
Acknowledgements
I would like to thank my primary supervisor Associate Professor Jayantha

Kodikara for his assistance and perseverance which enabled me to maintain the
focus required to pursue my research. Specific thanks also go to my associate
supervisors Dr David Marlow, Dr Paul Davis, Professor Stewart Burn and
Professor Xiao-Ling Zhao for their assistance and guidance, without which
this thesis could not have been assembled.
I greatly appreciate the help of the administrative and technical staff at the
Civil Engineering Department at Monash University, including but definitely
not limited to Jenny Manson, Chris Powell, Jane Moodie, Irene Sgouras, Noi
Souvandy and Long Goh.
I would also like to thank my colleagues at Monash University; Nathan
Rajeev, Nurses Kurucuk, Derek Chan, William Darlington, Mohan Yellishetty
and Ben Shannon for their help and friendship.
Thanks are also extended to my colleagues at CSIRO, specifically Steven
Cook, Fanny Boulaire and Grace Tjandraatmadja for their assistance and
teaching which allowed me to quickly gain vital new skills which were essential
to this project.
The author acknowledges the Australian Research Council which supported
this project via an ARC-Linkage Grant (LP 100200441). This work was con-
ducted in co-operation with City West Water Limited, CSIRO, Jemena, SP
AusNet, Envestra Limited, Water Corporation, Ipswich Water, South East
Water Limited and Queen’s University (Canada) for their financial and in-
kind support through this project.
Finally I would like to acknowledge the support and love from my beautiful
wife, my family and my friends.
xi
List of publications
The following is a list of publications resulting from the research undertaken

for this degree:
Refereed journal papers

Gould, S., Kodikara, J., Rajeev, P., Zhao, X.-L. and Burn, S., 2011. A void
ratio - water content - net stress model for environmentally stabilised expan-
sive soils. Canadian Geotechnical Journal, 48(6): 867-877.
Gould, S., Boulaire, F., Burn, S., Kodikara, J. and Zhao, X.-L., 2011. Sea-
sonal factors influencing the failure of buried water reticulation pipes. Water
Science and Technology - WST, 63(11): 2692-2699.
Gould, S., Rajeev, P., Kodikara, J., Zhao, X.-L., Burn, S. and Marlow, D.,
Under Review 2011. A new method for developing equations applied to the
water retention curve. Soil Science Society of America Journal.
Refereed conference papers

Boulaire, F., Gould, S., Beale, D., Kodikara, J., Burn, S. and Marlow, D.,
2010. Water Pipe Failure Predictions under Different Climate Scenarios, Cli-
mate Adaptation Futures: Preparing for the unavoidable impacts of climate
change, 29th June-1st July, Gold Coast, Australia.
Gould, S., Boulaire, F. and Kodikara, J., 2009. Understanding how the Aus-
tralian Climate can Affect Pipe Failure. Proceedings of OzWater 09. AWA,
16th-18th March, Melbourne, Australia.
xiii
xiv
Boulaire, F., Gould, S., Moglia, M. and Marlow, D., 2009. Integrating the
impact of climate into event based failure models for water pipes. Proceedings
of OzWater 09. AWA, 16th-18th March, Melbourne, Australia.
Gallage, C., Chan, D., Gould, S. and Kodikara, J. K., 2009. Stress-Strain
Development Of An In-Service Cast Iron Water Reticulation Pipe Buried In
Expansive Soil. Proceedings of OzWater 09. AWA, 16th-18th March, Mel-
bourne, Australia.
Chan, D., Gallage, C., Gould, S., Kodikara, J., Bouazza, A. and Cull, J.,
2009. Field Instrumentation Of Water Reticulation Pipe Buried In Reactive
Soil. OzWater 09. AWA, 16th-18th March, Melbourne, Australia.
Chan, D., Kodikara, J., Gould, S., Ranjith, P., Choi, X. S. K. and Davis,
P., 2007. Data analysis and laboratory investigation of the behaviour of pipes
buried in reactive clay. Proceedings of 10th Australia New Zealand Confer-
ence on Geomechanics - Common Ground 2007. 21st-24th October, Brisbane,
Australia.
Non-refereed conference papers

Gould, S. and Kodikara, J., 2008 Understanding how the Australian climate
affects pipe failure, Asia-Oceania Top University League on Engineering Post-
graduate conference, 26th-27th November, Auckland, New Zealand.
Research reports
Chan, D., Rajeev, P., Gould, S. and Kodikara, J., 2010. Field measurement
of the behaviour of an in-service water reticulation pipe buried in reactive soil
(Altona North, VIC) - Part 2. Research report, Monash University.
Gould, S. and Kodikara, J. K., 2009. Exploratory Statistical Analysis of Gas

Reticulation Main Failures (Melbourne, Australia). Technical Report RR12.
Gould, S. and Kodikara, J. K., 2008. Exploratory Statistical Analysis of

LIST OF PUBLICATIONS xv
Water Reticulation Main Failures (Melbourne, Australia). Technical Report

RR11.
Gallage, C., Chan, D., Gould, S. and Kodikara, J., 2009. Field measurement
of the behaviour of an in-service gas reticulation pipe buried in reactive soil
(Fawkner, VIC). Research report, Monash University.
Gallage, C., Chan, D., Gould, S. and Kodikara, J., 2008. Field measurement
of the behaviour of an in-service water reticulation pipe buried in reactive soil
(Altona North, VIC). Research report, Monash University.
Research posters
Gould, S., and Boulaire, F., 2009. Buried pipe failure prediction incorporat-
ing climate, Monash University HDR poster exhibition.
Gould, S., Gallage, C., Chan, D. and Kodikara, J., 2008. Understanding how
climate influences buried pipes - a field study, Monash University HDR poster
exhibition. Winner of the ’Outstanding Contribution Award’ for the best
poster from the Engineering Faculty.
Chan, D., Gallage, C., Gould, S., Amarasiri, A. and Kodikara, J., 2008. Field
instrumentation of gas pipe buried in reactive soil.
Nomenclature
• αer - fitting parameter • φ - fitting parameter
• αa - fitting parameter • φae - fitting parameter represent-

ing the curvature of the SWCC
• αf - field factor at ψae
• αω - coefficient of hydric soil ex- • φr - fitting parameter represent-

pansion (m/m) ing the curvature of the SWCC
at ψr
• ∆K - change in applied stress in-
√
tensity factor (M P a m) • ψ - suction (kP a)
• ∆σgb - applied flexural stress • ψ ∗ - log10 (ψ)

(M P a) • ψae - suction at air-entry value
• ∆σ0 - applied axial stress (kP a)
(M P a) • ψae
∗
- log10 (ψae )
• γ - soil unit weight (kN/m3 ) • ψmin - minimum suction value

(kP a)
• µ - mean of Poisson distribution
• ψmin
∗
- log10 (ψmin )
• ν - Poisson’s ratio for soil
• ψr - suction value at residual wa-
• ω - gravimetric water content ter content (kP a)
(g/g)
• ψr∗ - log10 (ψr )
• ωae - gravimetric water content
• σ - stress (M P a)
at air-entry value (g/g)
• σ0 - nominal stress (kP a)
• ωr - residual gravimetric water
content (g/g) • σ0 - pipe stress (P a)
• σ1 - pipe stress (P a)
• ωsat - saturated gravimetric wa-
ter content (g/g) • σ2 - pipe stress (P a)
xvii
xviii
• σ3 - pipe stress (P a) • d - notch, crack or pit depth (m)
• σgb - pipe stress (P a) • D - soil depth above the pipe

(m)
• σp - prism load (kP a)
• Di - pipe inside diameter (m)
• Θ - normalised volumetric water
content • Do - pipe outside diameter (m)
• e - soil void ratio

• a - fitting parameter
• er0 - fitting parameter
• A - cross-sectional area (m2 )
• Ec - soil modulus
• a0 - fitting parameter
• Es - secant modulus (GPa)
• a1 - constant
• er - fitting parameter
• a2 - constant
• er - void ratio of soil when ω = 0
• a3 - constant
• es - void ratio of soil when satu-
• a4 - constant rated, e = Gs .ω
• a5 - constant • F - tensile force (N)
• a6 - constant • F1 - calculation parameter
• a7 - constant • F2 - calculation parameter
• A0 - calculation parameter • Fb - geometric correction factor
• A1 - calculation parameter • Gs - soil specific gravity
• h - layer height (m)

• A2 - calculation parameter
• H - total height beneath pipe
• b - fitting parameter
(m)
• b - pipe diameter (m)
• i - time step
• b1 - constant
• i0 - geometric correction factor
• b2 - constant • i1 - geometric correction factor
• Bd - trench width (m) • i2 - geometric correction factor
• Cd - load coefficient • i3 - geometric correction factor
• cr - corrosion rate (mm/year) • If - depth factor

NOMENCLATURE xix
• Is - shape factor • Pi - internal pressure
• j - soil layer • R2 - coefficient of determination
• KI - applied stress intensity fac- • S1 initial slope of the SWCC

√
tor (M P a m)
• S2 centre final slope of the
• KIC - fracture toughness SWCC
√
(M P a m)
• S3 final slope of the SWCC
• ks - soil stiffness (kP a)
• Sr - degree of saturation (%)
• L - length of modelled section
(m) • t - wall thickness (m)
• m0 - calculation parameter • t1 - fitting parameter

• m - fitting parameter • t2 - fitting parameter
• n0 - calculation parameter
• W - specimen width (m)
• n - fitting parameter
• Wt - maximum Marston load per
• o - fitting parameter unit length
Contents
Dedication iii
Copyright Notices v
Declaration vii
Executive Summary ix
Acknowledgements xi
List of publications xiii
Nomenclature xvii
Contents xxi
List of Figures xxix
List of Tables xxxvii
1 Thesis introduction 1
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Aims and scope of research . . . . . . . . . . . . . . . . . . . . . 3
1.3 Thesis overview . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2 Factors affecting buried pipe failure 7

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 Defining pipe failure . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 Pipe structural capacity and its deterioration . . . . . . . . . . . 8
2.4 Sources of buried pipe loading . . . . . . . . . . . . . . . . . . . 9
2.4.1 Ordinary loads . . . . . . . . . . . . . . . . . . . . . . . 10
2.4.1.1 Internal pressure . . . . . . . . . . . . . . . . . 10
xxi
xxii
2.4.1.2 Dead load . . . . . . . . . . . . . . . . . . . . . 11

2.4.1.3 Live load . . . . . . . . . . . . . . . . . . . . . 12
2.4.2 Non-service loads . . . . . . . . . . . . . . . . . . . . . . 12
2.4.2.1 Soil shrinkage/swelling . . . . . . . . . . . . . . 13
2.4.2.2 Loss of support . . . . . . . . . . . . . . . . . . 14
2.4.2.3 Frost loading . . . . . . . . . . . . . . . . . . . 14
2.4.2.4 Surge pressure . . . . . . . . . . . . . . . . . . 15
2.4.2.5 Restrained thermal contraction . . . . . . . . . 15
2.5 Types of pipe failure . . . . . . . . . . . . . . . . . . . . . . . . 15
2.5.1 Circumferential failure . . . . . . . . . . . . . . . . . . . 16
2.5.2 Longitudinal failure . . . . . . . . . . . . . . . . . . . . . 16
2.5.3 Split bell . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.5.4 Blown section . . . . . . . . . . . . . . . . . . . . . . . . 17
2.5.5 Through-wall defects . . . . . . . . . . . . . . . . . . . . 17
2.5.6 Third-party damage . . . . . . . . . . . . . . . . . . . . 18
2.5.7 Tapping failure . . . . . . . . . . . . . . . . . . . . . . . 18
2.6 Factors affecting pipe failure rate . . . . . . . . . . . . . . . . . 21
2.6.1 Pipe material . . . . . . . . . . . . . . . . . . . . . . . . 21
2.6.2 Pipe diameter . . . . . . . . . . . . . . . . . . . . . . . . 22
2.6.3 Surrounding soil . . . . . . . . . . . . . . . . . . . . . . . 23
2.6.4 Pipe age . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.6.5 Season . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
2.6.5.1 Winter . . . . . . . . . . . . . . . . . . . . . . . 25
2.6.5.2 Summer . . . . . . . . . . . . . . . . . . . . . . 27
2.7 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3 Pipe network historical failure analysis 29

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.2 Network data . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
3.3 Data preparation . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.4 Exploratory statistical investigation . . . . . . . . . . . . . . . . 34
3.4.1 Calculation of failure rates . . . . . . . . . . . . . . . . . 34
3.4.1.1 Recording rates . . . . . . . . . . . . . . . . . . 34
3.4.1.2 Calculation of the exposed length . . . . . . . . 34
3.4.1.3 Calculation of failure rates . . . . . . . . . . . . 35
3.4.1.3.1 Yearly failure rates . . . . . . . . . . . 35
3.4.2 Pipe material . . . . . . . . . . . . . . . . . . . . . . . . 35
3.4.3 Pipe diameter . . . . . . . . . . . . . . . . . . . . . . . . 37
CONTENTS xxiii
3.4.4 Pipe material - pipe diameter interaction . . . . . . . . . 37

3.4.5 Soil type . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
3.4.6 Pipe material - soil type interaction . . . . . . . . . . . . 44
3.4.7 Pipe diameter - soil type interaction . . . . . . . . . . . 47
3.4.8 Failure type . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.4.9 Pipe age . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
3.4.10 Pipe age - pipe material interaction . . . . . . . . . . . . 50
3.5 Investigation of seasonal factors influencing failure rate . . . . . 52
3.5.1 Intra-year variation of failure rates . . . . . . . . . . . . 52
3.5.2 Climatic parameter correlation investigation . . . . . . . 52
3.5.3 Interaction between climate and other factors . . . . . . 56
3.5.3.1 Effect of climate and pipe material on failure . 56
3.5.3.2 Effect of climate and pipe diameter on failure . 58
3.5.3.3 Effect of climate and different soil types on failure 59
3.5.3.4 Effect of climate on observed failure type . . . . 60
3.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
4 Field instrumentation of an in-service pipe in a high shrink/swell

potential soil 63
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
4.2 Site selection and investigation . . . . . . . . . . . . . . . . . . . 64
4.2.1 Site selection criteria . . . . . . . . . . . . . . . . . . . . 64
4.2.2 Site soil bore hole analysis . . . . . . . . . . . . . . . . . 65
4.2.3 Site geometry . . . . . . . . . . . . . . . . . . . . . . . . 65
4.2.4 Soil classification tests . . . . . . . . . . . . . . . . . . . 66
4.2.4.1 Particle size analysis . . . . . . . . . . . . . . . 66
4.2.4.2 Plastic and liquid limits . . . . . . . . . . . . . 67
4.2.4.3 Plasticity index . . . . . . . . . . . . . . . . . . 68
4.2.4.4 Linear shrinkage . . . . . . . . . . . . . . . . . 68
4.2.4.5 Specific gravity, dry density and void ratio . . . 69
4.2.4.6 Swelling properties . . . . . . . . . . . . . . . . 69
4.2.4.7 Mineralogy analysis . . . . . . . . . . . . . . . 71
4.2.4.8 Summary . . . . . . . . . . . . . . . . . . . . . 72
4.3 Sensors and instrumentation equipment . . . . . . . . . . . . . . 73
4.3.1 Strain gauges . . . . . . . . . . . . . . . . . . . . . . . . 74
4.3.2 Earth pressure cells . . . . . . . . . . . . . . . . . . . . . 74
4.3.3 Thermocouples . . . . . . . . . . . . . . . . . . . . . . . 74
4.3.4 Matric suction sensors . . . . . . . . . . . . . . . . . . . 74
xxiv
4.3.4.1 Calibration of matric suction sensors . . . . . . 75

4.3.5 Soil water sensors . . . . . . . . . . . . . . . . . . . . . . 75
4.3.5.1 Calibration of soil water sensors . . . . . . . . . 75
4.3.6 Pipe water pressure and temperature gauges . . . . . . . 76
4.3.7 Rod extensometer . . . . . . . . . . . . . . . . . . . . . . 77
4.3.8 Weather station . . . . . . . . . . . . . . . . . . . . . . . 78
4.3.9 Neutron probe . . . . . . . . . . . . . . . . . . . . . . . . 79
4.3.10 Data acquisition system . . . . . . . . . . . . . . . . . . 79
4.4 Field work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
4.4.1 Excavation . . . . . . . . . . . . . . . . . . . . . . . . . . 87
4.4.2 Sensor installation (7th - 14th of January 2008) . . . . . . 87
4.4.2.1 Strain gauges . . . . . . . . . . . . . . . . . . . 87
4.4.2.2 Earth pressure cells . . . . . . . . . . . . . . . . 89
4.4.2.3 Soil monitoring sensors . . . . . . . . . . . . . . 90
4.4.2.4 Pipe water pressure and temperature gauges . . 91
4.4.2.5 Wiring . . . . . . . . . . . . . . . . . . . . . . . 91
4.4.3 Backfilling . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.4.4 Supplementary sensor installations . . . . . . . . . . . . 93
4.4.4.1 Weather station (19th of February 2008) . . . . 93
4.4.4.2 Rod extensometer (4th of March 2008) . . . . . 93
4.4.4.3 Neutron probe (3rd of February 2009) . . . . . . 94
4.4.5 Data Acquisition Timetable . . . . . . . . . . . . . . . . 95
4.5 Investigation of field data . . . . . . . . . . . . . . . . . . . . . 98
4.5.1 Pipe stress . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.5.2 Correlation of flexural stress with soil shrinkage . . . . . 101
4.5.3 Pipe-soil behaviour . . . . . . . . . . . . . . . . . . . . . 102
4.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103
5 Mechanical properties of Australian Cast Iron reticulation pipes105

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
5.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
5.3 Previous work . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5.4 Material deterioration . . . . . . . . . . . . . . . . . . . . . . . 109
5.5 Experimental Work . . . . . . . . . . . . . . . . . . . . . . . . . 109
5.5.1 Sample Details . . . . . . . . . . . . . . . . . . . . . . . 109
5.5.2 Tensile strength and secant modulus . . . . . . . . . . . 109
5.5.3 Fracture toughness . . . . . . . . . . . . . . . . . . . . . 110
5.5.4 Fatigue . . . . . . . . . . . . . . . . . . . . . . . . . . . . 112
CONTENTS xxv
5.6 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . 112

5.6.1 Tensile strength . . . . . . . . . . . . . . . . . . . . . . . 112
5.6.2 Secant modulus . . . . . . . . . . . . . . . . . . . . . . . 116
5.6.3 Fracture toughness . . . . . . . . . . . . . . . . . . . . . 117
5.6.4 Fatigue . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
5.7 Fatigue failure prediction . . . . . . . . . . . . . . . . . . . . . . 120
5.8 Failure mode comparison . . . . . . . . . . . . . . . . . . . . . . 122
5.9 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
6 A novel technique for model development 127

6.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
6.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
6.3 Equation development technique . . . . . . . . . . . . . . . . . . 131
6.3.1 Plotting the derivative of the SWCC . . . . . . . . . . . 131
6.3.2 Equation development . . . . . . . . . . . . . . . . . . . 132
6.3.3 Updated equation . . . . . . . . . . . . . . . . . . . . . . 135
6.3.4 Fitting parameter effect study . . . . . . . . . . . . . . . 136
6.3.5 Equation and derivative from Pham and Fredlund (2008) 141
6.4 Evaluation of SWCC equation using experimental data . . . . . 142
6.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
6.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 152
7 A void ratio - water content - net stress model for environ-

mentally stabilised expansive soils 155
7.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
7.2 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
7.2.1 Previous work . . . . . . . . . . . . . . . . . . . . . . . . 159
7.3 Shrinkage curve equation definition . . . . . . . . . . . . . . . . 160
7.3.1 Parameter affect study . . . . . . . . . . . . . . . . . . . 161
7.3.2 Evaluation e − ω equation using published data . . . . . 163
7.3.3 Comparison to other equations . . . . . . . . . . . . . . 163
7.3.4 Comparison of automated fitted results to manual fitting
of the e − ω equation . . . . . . . . . . . . . . . . . . . . 163
7.4 Shrinkage surface equation definition . . . . . . . . . . . . . . . 174
7.4.1 Evaluation of e − ω − σ equation using published data . 175
7.4.2 Use of compression index in the constant e−ω −σ equation175
7.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
xxvi
7.5.1 Theoretical understanding of fitting parameters in the

e − ω equation . . . . . . . . . . . . . . . . . . . . . . . 181
7.5.2 Theoretical understanding of fitting parameters in the
e − ω − σ equation . . . . . . . . . . . . . . . . . . . . . 183
7.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
8 Modelling pipe-soil interaction 185

8.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 185
8.2 Numerical modelling options . . . . . . . . . . . . . . . . . . . . 185
8.3 Theoretical framework for modelling . . . . . . . . . . . . . . . 186
8.4 Model description . . . . . . . . . . . . . . . . . . . . . . . . . . 190
8.4.1 Calculation of soil stiffness, ks . . . . . . . . . . . . . . . 192
8.4.2 Calculation of soil hydric expansion coefficient, ασ . . . 194
8.4.3 Calculation of change in soil height at constant stress, hω 194
8.4.4 Numerical Winkler model . . . . . . . . . . . . . . . . . 194
8.5 Application and validation of the model . . . . . . . . . . . . . 195
8.5.1 Pipe mechanical properties . . . . . . . . . . . . . . . . . 196
8.5.2 Field site e − ω − σ soil surface . . . . . . . . . . . . . . 196
8.5.3 Modelling software . . . . . . . . . . . . . . . . . . . . . 199
8.5.4 Modelling results for fixed end support conditions . . . . 200
8.5.5 Modelling results for pinned end support conditions . . . 202
8.5.6 Averaged modelling results . . . . . . . . . . . . . . . . . 204
8.6 Parametric study of the model . . . . . . . . . . . . . . . . . . . 205
8.6.1 The effect of changing soil stiffness and hydric expansion
coefficient . . . . . . . . . . . . . . . . . . . . . . . . . . 207
8.6.2 Effect of soil water content change . . . . . . . . . . . . . 208
8.6.3 Effect of changing the uniformly distributed dead load . 210
8.7 Pipe lifetime estimation . . . . . . . . . . . . . . . . . . . . . . 211
8.8 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213
9 Conclusions and recommendations 215

9.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
9.2 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 215
9.3 Recommendations for future work . . . . . . . . . . . . . . . . . 217
9.3.1 Pipe joint characterisation . . . . . . . . . . . . . . . . . 217
9.3.2 End support characterisation . . . . . . . . . . . . . . . 217
9.3.3 Validation of the e − ω − σ model . . . . . . . . . . . . . 217
9.3.4 Continuum modelling . . . . . . . . . . . . . . . . . . . . 218
CONTENTS xxvii
9.3.5 Asset management application . . . . . . . . . . . . . . . 218
10 References 219
A Pipe network failure analysis data preparation 233

A.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 233
A.2 Asset data cleaning . . . . . . . . . . . . . . . . . . . . . . . . . 233
A.2.1 Repeated asset ID test . . . . . . . . . . . . . . . . . . . 234
A.2.1.1 No construction date test . . . . . . . . . . . . 235
A.2.1.2 No material type test . . . . . . . . . . . . . . . 235
A.2.1.3 No or short length test . . . . . . . . . . . . . . 236
A.2.1.4 No diameter test . . . . . . . . . . . . . . . . . 236
A.2.1.5 No soil type test . . . . . . . . . . . . . . . . . 237
A.2.1.6 Large diameter test . . . . . . . . . . . . . . . . 237
A.2.1.7 Construction date in future test . . . . . . . . . 237
A.2.1.8 Unlikely material type test . . . . . . . . . . . . 237
A.3 Failure data cleaning . . . . . . . . . . . . . . . . . . . . . . . . 238
A.3.1 Repeated failure ID test . . . . . . . . . . . . . . . . . . 238
A.3.2 No failure date test . . . . . . . . . . . . . . . . . . . . . 239
A.3.3 Failure date in future test . . . . . . . . . . . . . . . . . 239
A.3.4 Matching test . . . . . . . . . . . . . . . . . . . . . . . . 239
A.3.4.1 Matched to erroneous asset test . . . . . . . . . 240
A.3.5 Preconstruction failure date test . . . . . . . . . . . . . . 240
A.3.6 Merging of CWW and SEWL data . . . . . . . . . . . . 240
A.4 Data grouping . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
A.4.1 Grouping of materials . . . . . . . . . . . . . . . . . . . 241
A.4.2 Grouping of diameters . . . . . . . . . . . . . . . . . . . 242
A.4.3 Grouping of failure types . . . . . . . . . . . . . . . . . . 243
A.5 Soil data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
B Field data 247

B.1 Earth pressure . . . . . . . . . . . . . . . . . . . . . . . . . . . . 247
B.2 Pipe water pressure . . . . . . . . . . . . . . . . . . . . . . . . . 248
B.3 Soil temperature . . . . . . . . . . . . . . . . . . . . . . . . . . 250
B.4 Soil volumetric water content and suction . . . . . . . . . . . . . 253
B.5 Weather station . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
B.6 Soil movement . . . . . . . . . . . . . . . . . . . . . . . . . . . . 258
B.7 Pipe strain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
xxviii
C Soil shrinkage curve equation development 263

C.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
List of Figures
2.1 Manufacturing flaw in wall of PVC pipe . . . . . . . . . . . . . 9

2.2 Schematic representation of ordinary loads acting on buried pipe 10
2.3 Schematic representation of common non-service loads acting
on buried pipe . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.4 Circumferential pipe failure . . . . . . . . . . . . . . . . . . . . 19
2.5 Longitudinal pipe failure . . . . . . . . . . . . . . . . . . . . . . 20
2.6 Examples of other failure types . . . . . . . . . . . . . . . . . . 20
3.1 CWW and SEWL Assets and Supply Boundaries . . . . . . . . 31

3.2 Length of assets installed in each year . . . . . . . . . . . . . . . 33
3.3 Annual failure rates over the observation period († Adjusted fail-
ure rate) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
3.4 Average failure rate by materials over observation period . . . . 38
3.5 Average Failure rate by diameter over observation period . . . . 40
3.6 Length of assets in each diameter/material group . . . . . . . . 41
3.7 Failure rates for assets in each diameter/material group . . . . . 42
3.8 Average failure rate by soil type over observation period . . . . 43
3.9 Updated soil classifications (ME and EX soil types combined) . 44
3.10 Length of assets in each soil type and material group . . . . . . 45
3.11 Failure rates for assets in each soil and material group . . . . . . 47
3.12 Length of assets in each Soil Type and Diameter Group . . . . . 48
3.13 Failure rates for assets in each Soil Type and Diameter Group . 49
3.14 Failure rate of each failure type over the observation period . . . 50
3.15 Failure rates of assets as they age . . . . . . . . . . . . . . . . . 51
3.16 Failure rates of assets as they age by material . . . . . . . . . . 51
3.17 Intra-year failure rate variation . . . . . . . . . . . . . . . . . . 53
3.18 Histogram of pipe failure rate depending on Minimum API . . . 55
3.19 Intra-year failure rate variation by Material . . . . . . . . . . . 58
3.20 Intra-year failure rate variation by Diameter . . . . . . . . . . . 59
xxix
xxx
3.21 Intra-year failure rate variation by Soil . . . . . . . . . . . . . . 60

3.22 Intra-year failure rate variation by Failure Type . . . . . . . . . 61
4.1 Geological map of the CWW supply area showing the location
of the Altona North field instrumentation site Rixon (1973) . . . 66
4.2 Bore hole log data from the Altona North field instrumentation
site (Note: the ground water table was not observed during
sampling) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
4.3 Particle size distribution of soil sample from Altona North field
study site . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
4.4 Dry density and void ratio of undisturbed soil collected from
field site . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 70
4.5 Oedometers used for determining shrink/swell properties . . . . 71
4.6 Soil swelling strain vs. applied pressure for a sample from field
instrumentation site . . . . . . . . . . . . . . . . . . . . . . . . 72
4.7 Calibration of suction sensors with pressure cell . . . . . . . . . 76
4.8 Calibration of suction sensors . . . . . . . . . . . . . . . . . . . 77
4.9 Calibration of soil water sensors . . . . . . . . . . . . . . . . . . 78
4.10 Rod-extensometer and the location of each installed anchor . . . 79
4.11 Schematic of neutron probe . . . . . . . . . . . . . . . . . . . . 80
4.12 Calibration of neutron probe . . . . . . . . . . . . . . . . . . . . 81
4.13 The data acquisition system installed at the site . . . . . . . . . 82
4.14 A detailed plan of the Altona North instrumentation site . . . . 84
4.15 Vertical cross-section and labelling of sensors in instrumentation
pits - part A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
4.16 Vertical cross-section labelling of sensors in instrumentation pits
- part B . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.17 Schematic of a typical instrumentation pit (plan view) . . . . . 87
4.18 Excavation for the instrumentation . . . . . . . . . . . . . . . . 88
4.19 Strain gauging on the pipe . . . . . . . . . . . . . . . . . . . . . 89
4.20 Installation of an earth pressure cell . . . . . . . . . . . . . . . . 90
4.21 Horizontal drilling in preparation of sensor installation . . . . . 91
4.22 Installation of a thermocouple . . . . . . . . . . . . . . . . . . . 92
4.23 Installation of a suction sensor . . . . . . . . . . . . . . . . . . . 92
4.24 Installation of a soil water sensors . . . . . . . . . . . . . . . . . 93
4.25 Pipe water temperature and pressure gauges . . . . . . . . . . . 94
4.26 Wiring after installation of sensors . . . . . . . . . . . . . . . . 95
4.27 Backfilling and restoration of the site . . . . . . . . . . . . . . . 96
LIST OF FIGURES xxxi
4.28 Weather station after installation at the site . . . . . . . . . . . 97

4.29 Installation of access tube for neutron probe . . . . . . . . . . . 97
4.30 Expected trends in pipe stress and soil water content . . . . . . 99
4.31 Average axial stress in pipe . . . . . . . . . . . . . . . . . . . . 100
4.32 Flexural stress in pipe . . . . . . . . . . . . . . . . . . . . . . . 101
4.33 Change in height of soil beneath pipe (measure by rod exten-
someter) with average pipe stress . . . . . . . . . . . . . . . . . 102
4.34 Predicted vertical pipe movement in winter and summer based
on the stress analysis . . . . . . . . . . . . . . . . . . . . . . . . 103
5.1 Tensile specimen dimensions . . . . . . . . . . . . . . . . . . . . 110

5.2 Tensile specimen dimensions . . . . . . . . . . . . . . . . . . . . 111
5.3 DENT specimen dimensions . . . . . . . . . . . . . . . . . . . . 112
5.4 Stress-Strain curve for Specimens C2-T1 (Spun Cast) and S2-T6
(Pit Cast) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
5.5 Uniaxial Tensile Strength of Grey Cast Iron against Sample Age 114
5.6 Correlation between graphitisation depth to thickness ratio and
tensile strength . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
5.7 Comparison of results from UK, Japan, Canada and Australia
(only trendlines shown) . . . . . . . . . . . . . . . . . . . . . . . 116
5.8 Secant Modulus . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
5.9 Fatigue Test Results . . . . . . . . . . . . . . . . . . . . . . . . 120
5.10 Log-normal distribution fitted for pit geometry . . . . . . . . . . 121
5.11 Failure Mode Comparison . . . . . . . . . . . . . . . . . . . . . 124
5.12 Failure Mode Comparison . . . . . . . . . . . . . . . . . . . . . 125
6.1 Idealised SWCC of volume change soil with labelled features

(x-axis in log scale) . . . . . . . . . . . . . . . . . . . . . . . . . 128
6.2 Idealised derivative of the soil water characteristic curve (x-axis
in log scale) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132
6.3 Parameter sensitivity results (x-axis in log scale). Note: Except
∗
where otherwise stated ψae = 101.5 = 32 kPa (ψae = 1.5), ψr =
103.5 = 3162 kPa, m = −0.025, n = −0.45, ωsat = 1.25, ψmin =
100 = 1 kPa, φae = 5 and φr = 5. . . . . . . . . . . . . . . . . . . 137
∗
103.5 = 3162 kPa, m = −0.025, n = −0.45, ωsat = 1.25, ψmin =
100 = 1 kPa, φae = 5 and φr = 5. . . . . . . . . . . . . . . . . . . 138
xxxii

∗
3.5
10 = 3162 kPa, m = −0.025, n = −0.45, ωsat = 1.25, ψmin =
100 = 1 kPa, φae = 5 and φr = 5. . . . . . . . . . . . . . . . . . . 139
∗
3.5
10 = 3162 kPa, m = −0.025, n = −0.45, ωsat = 1.25, ψmin =
100 = 1 kPa, φae = 5 and φr = 5. . . . . . . . . . . . . . . . . . . 140
6.7 Silty Loam (Raw data from University of Saskatchewan, per-
sonnel communication S.L. Barbour (2004) (x-axis in log scale) . 144
6.8 Silty Clay (Raw data from University of Saskatchewan, person-
nel communication S.L. Barbour (2004) (x-axis in log scale) . . 145
6.9 Sandy Loam (Raw data from University of Saskatchewan, per-
sonnel communication S.L. Barbour (2004) (x-axis in log scale) . 146
6.10 Regina Clay (Raw data from Fredlund and Xing (1994)) (x-axis
in log scale) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147
6.11 Kidd Creek Tailings (Raw data from Fredlund and Xing (1994))
(x-axis in log scale) . . . . . . . . . . . . . . . . . . . . . . . . . 148
6.12 Jossigny Loam (Raw data from Fleureau et al. (2002)) (x-axis
in log scale) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
6.13 Field site soil from ground level to 350 mm depth (Raw data
from Chan (InPrep)) (x-axis in log scale) . . . . . . . . . . . . . 150
6.14 Field site soil from 350 mm to 500 mm depth (Raw data from
Chan (InPrep)) (x-axis in log scale) . . . . . . . . . . . . . . . . 151
7.1 Comparison of Water Content and Suction Curves. Adapted

from Fleureau et al. (2002) . . . . . . . . . . . . . . . . . . . . . 158
7.2 Idealised shrinkage curve . . . . . . . . . . . . . . . . . . . . . . 159
7.3 Parameter Sensitivity. Note: Except where otherwise stated
a = 0.8, b = 0.2, er = 1.0, φ = 200 and m = 2.7. (a) the effect of
changing fitting parameter a, (b) the effect of changing fitting
parameter b, (c) the effect of changing fitting parameter er , (d)
the effect of changing fitting parameter φ and (e) the effect of
changing fitting parameter m . . . . . . . . . . . . . . . . . . . 162
7.4 e − ω equation fit to Olsen and Haugen (1998) . . . . . . . . . . 166
7.5 e − ω equation fit to Peng and Horn (2005) . . . . . . . . . . . . 167
7.6 e − ω equation fit to Reeve and Hall (1978) (soils 1 to 3 of 6) . . 168
7.7 e − ω equation fit to Reeve and Hall (1978) (soils 4 to 6 of 6) . . 169
LIST OF FIGURES xxxiii
7.8 e − ω equation fit to Talsma (1977) . . . . . . . . . . . . . . . . 170

7.9 e − ω equation fit to Tripathy et al. (2002) soil A . . . . . . . . 171
7.10 e − ω equation fit to Tripathy et al. (2002) soil B . . . . . . . . 172
7.11 Automated vs. Manual fitting results . . . . . . . . . . . . . . . 173
7.12 e − ω − σ equation surface fit to data from Tripathy et al. (2002)
Soil A . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
7.13 e − ω − σ equation constant e contours fit to data from Tripathy
et al. (2002) Soil A . . . . . . . . . . . . . . . . . . . . . . . . . 177
7.14 e − ω − σ equation constant σ contours fit to data from Tripathy
et al. (2002) Soil A . . . . . . . . . . . . . . . . . . . . . . . . . 178
7.15 e−ω −σ equation constant ω contours fit to data from Tripathy
et al. (2002) Soil A . . . . . . . . . . . . . . . . . . . . . . . . . 179
7.16 Partial derivatives of surface fit to data from Tripathy et al.
(2002) Soil A. (a) Partial derivatives with respect to ω at a
range of σ’s (b) Partial derivatives with respect σ to at a range
of ω’s . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
8.1 Idealised e − ω − σ surface . . . . . . . . . . . . . . . . . . . . . 187

8.2 Stress attenuation with depth . . . . . . . . . . . . . . . . . . . 189
8.3 Flowchart of modelling procedure . . . . . . . . . . . . . . . . . 192
8.4 Combination of soil layer to single representative spring . . . . . 193
8.5 Schematic of spring model . . . . . . . . . . . . . . . . . . . . . 195
8.6 Comparison of estimated unrestrained height change and rod
extensometer measurements . . . . . . . . . . . . . . . . . . . . 197
8.7 e − ω − σ surface fit for the Altona North field study site . . . . 198
8.8 Soil water content data from the Altona North field study site . 199
8.9 Modelling estimations for fixed end support conditions . . . . . 201
8.10 Comparison of the change in pipe flexural stress estimated for
fixed end support conditions and measurements from the Altona
North field study site . . . . . . . . . . . . . . . . . . . . . . . . 202
8.11 Modelling estimations for pinned end support conditions . . . . 203
8.12 Comparison of the change in pipe flexural stress estimated for
pinned end support conditions and measurements from the Al-
tona North field study site . . . . . . . . . . . . . . . . . . . . . 204
8.13 Comparison of the change in the pipe flexural stresses averaged
between pinned and fixed end support conditions, and measure-
ments from the Altona North field study site . . . . . . . . . . . 205
8.14 Base line soil water content cycle . . . . . . . . . . . . . . . . . 206
xxxiv
8.15 Change in pipe flexural stress for three soil types and water
content changes against ks . . . . . . . . . . . . . . . . . . . . . . 208
8.16 Change in pipe flexural stress for three soil types and water
content changes against ασ . . . . . . . . . . . . . . . . . . . . . 209
8.17 Change in pipe flexural stress for 50%, 100% and 200% of the
base line magnitude . . . . . . . . . . . . . . . . . . . . . . . . . 210
8.18 Pipe lifetime estimation incorporating soil type and corrosion
rate for net section collapse failure . . . . . . . . . . . . . . . . 212
8.19 Pipe lifetime estimation incorporating soil type and corrosion
rate for crack growth failure . . . . . . . . . . . . . . . . . . . . 213
A.1 Potential difference between calculated and actual asset length . 236
A.2 Final soil classifications . . . . . . . . . . . . . . . . . . . . . . . 246
B.1 Response of earth pressure cells in Pits 2 and 3 part A . . . . . 248

B.2 Pipe water pressure . . . . . . . . . . . . . . . . . . . . . . . . . 249
B.3 Comparison of daily temperature at Pit 1 . . . . . . . . . . . . . 250
B.6 Comparison of daily temperature at Pit 3 away from pipe . . . . 252
B.7 Daily average soil volumetric water content in Pit 1 . . . . . . . 253
B.10 Daily average soil volumetric water content in Pit 3 away from
pipe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
B.11 Daily average soil suction in Pit 1 . . . . . . . . . . . . . . . . . 255
B.14 Daily average soil suction in Pit 3 away from pipe . . . . . . . . 257
B.15 Daily rainfall recorded by the weather station . . . . . . . . . . 258
B.16 Daily soil movement . . . . . . . . . . . . . . . . . . . . . . . . 259
B.17 Daily average longitudinal strain at Pit 1 . . . . . . . . . . . . . 260
B.20 Daily average circumferential strain at Pit 1 . . . . . . . . . . . 261
C.1 Idealised version of shrinkage curve . . . . . . . . . . . . . . . . 264

LIST OF FIGURES xxxv
C.2 Idealised derivative of the shrinkage curve . . . . . . . . . . . . 265

C.3 Modelled shrinkage curve, where e0 = 0.5, a = 0.6, b = 0.1,
m = 2.5, φ = 100 and Gs = 2.5 . . . . . . . . . . . . . . . . . . 266
C.4 Modelled shrinkage curve derivative, where e0 = 0.5, a = 0.6,
b = 0.1, m = 2.5, φ = 100 and Gs = 2.5 . . . . . . . . . . . . . . 267
List of Tables
2.1 Factors affecting pipe failure rate (Kleiner and Rajani, 2002) . . 21
3.1 CWW and SEWL supplied asset data summary . . . . . . . . . 30

3.2 CWW and SEWL supplied failure data summary . . . . . . . . 30
3.3 Updated asset data summary . . . . . . . . . . . . . . . . . . . 32
3.4 Updated failure data summary . . . . . . . . . . . . . . . . . . . 32
3.5 Asset data to analysis summary . . . . . . . . . . . . . . . . . . 33
3.6 Failure data to analysis summary . . . . . . . . . . . . . . . . . 33
3.7 Recording rates . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.8 Average failure rates over observation period . . . . . . . . . . . 36
3.9 Length (km) of assets in-service, shown by pipe material and
pipe diameter. The material/diameter groups with the greatest
length are highlighted . . . . . . . . . . . . . . . . . . . . . . . . 39
3.10 Soil codes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
3.11 Percentage of soil types within authority boundaries . . . . . . . 42
3.12 Percentage of soil types where pipes are present . . . . . . . . . 45
3.13 Length (km) of assets in-service, shown by pipe material and soil
type. The material/diameter groups with the greatest length
are highlighted . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
3.14 Assets in-service, shown by diameter (mm) and soil type by
length (km). The material/diameter groups with the greatest
length are highlighted extra line extra line extra line . . . . . . . 48
3.15 Exposed length and number of failures of material and diameters
included in analysis . . . . . . . . . . . . . . . . . . . . . . . . . 58
4.1 Mineralogical content of soil sample from field site . . . . . . . . 72

4.2 Summary of the soil classification test results . . . . . . . . . . . 73
4.3 Data acquisition timetable . . . . . . . . . . . . . . . . . . . . . 96
4.4 Investigated data timetable . . . . . . . . . . . . . . . . . . . . 99
xxxvii
xxxviii
5.1 Properties of uncorroded CI from previous studies . . . . . . . . 108

5.2 Example calculations for fatigue testing . . . . . . . . . . . . . . 112
5.3 Cast Iron installations periods (Scott, 1990) . . . . . . . . . . . 114
5.4 Fracture toughness test results . . . . . . . . . . . . . . . . . . . 118
5.5 Failure scenario calculation parameters . . . . . . . . . . . . . . 123
7.1 Comparison of shrinkage curve equations . . . . . . . . . . . . . 160

7.2 Soil Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
7.3 Comparison of equation using R2 . . . . . . . . . . . . . . . . . 165
8.1 Properties and dimensions of modelled cast iron pipe . . . . . . 196

8.2 Values of ks and ασ for different soil types . . . . . . . . . . . . 207
8.3 Constants for exponential corrosion model (Rajani et al., 2000) . 212
A.1 Asset Data Summary after Repeated Asset ID Test . . . . . . . 235

A.2 Material Introduction/Cessation Dates . . . . . . . . . . . . . . 238
A.3 Failure Data Summary after Repeated Failure ID Test . . . . . 239
A.4 WSAA pipeline Acronyms (Water Services Association of Aus-
tralia, 2007) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 241
A.5 Material group summary . . . . . . . . . . . . . . . . . . . . . . 242
A.6 Diameter summary . . . . . . . . . . . . . . . . . . . . . . . . . 242
A.7 Summary of diameters after grouping . . . . . . . . . . . . . . . 243
A.8 Summary of failure types after grouping . . . . . . . . . . . . . 243
A.9 Soil reclassification scheme for grant codes . . . . . . . . . . . . 245
A.10 Soil shrink/swell potential summary . . . . . . . . . . . . . . . . 246
Chapter 1
Thesis introduction
1.1 Introduction
Buried reticulation pipe networks are part of the critical infrastructure used
in every city in the developed world. These networks provide homes and
businesses with essential services, primarily the supply of potable water, and
other services such as the supply of natural gas and the removal of storm water
and sewerage. Within Australia alone 138,000 km of pipes currently supply
approximately 21.5 million people with water (National Water Commission
and WSAA, 2010). In most countries, these networks include pipes which are
over 100 years of age. In Melbourne, Australia the water reticulation pipe
network includes pipes which have been in use for over 130 years.
As the reticulation pipe networks continue to age and increasing numbers
of pipes exceed their design lives, the risk of failure increases. In fact the
failure of buried reticulation pipes is an undesirably common occurrence which
results in negative economic, social and environmental consequences. Some
potential consequences of pipe failures include costly and disruptive damage
to surrounding infrastructure such as roads and buildings, soil erosion and
landslides (e.g., the Thredbo landslide disaster in 1997 was caused by the
failure of a high pressure water pipe), service interruptions and traffic delays,
and contamination of storm water and associated waterways.
The simplest solution to this problem is to completely replace the existing
networks. However the capital cost of these networks makes this solution
prohibitively expensive. An estimate of the current written down replacement
cost of fixed water supply assets for the all Australian networks combined is
A$49.2 billion (National Water Commission and WSAA, 2010). This value is
not the true replacement cost, as it is the replacement cost less accumulated
1
2
depreciation. The true replacement cost is there for in excess of this value.
Additionally this estimate does not include the cost of replacement to the
community resulting from the loss of service, disruption of access (such as due
to road works) or potential damage to surrounding infrastructure. Therefore,
the private companies and public authorities responsible for operating and
maintaining these networks have opted to manage the pipe infrastructure with
targeted maintenance and replacement. In 2008/2009, the cost of maintaining
and replacing existing urban water assets in Australia was in excess of A$3.5
billion dollars (National Water Commission and WSAA, 2010).
The number of pipe failures expected to occur within a network is predicted

to inform strategic planning processes. The most commonly used method
identifies pipes using statistical modelling which is based on historical failure
records and pipe characteristics identified from pipe design principles. These
statistical models require calibration and once calibrated are location specific
and cannot be applied to other networks. In addition they normally consider
only pipe characteristics and do not consider the effect of the environment
on failure rates within and between years. Physical models also exist for the
prediction of pipe failure. These models are based on the well known relation-
ships between failure type and load configuration. These models are limited to
specific pipe materials and failure types, predict failure resulting from mate-
rial deterioration and account for loading due to soil weight, internal pressure
and/or traffic. These existing models do not consider variations in loading
conditions as the result of changing environmental conditions over the course
of the calendar year.
In the papers that have examined seasonal variation in failure rates, the
failure of buried reticulation pipes has been reported to occur at different
rates at different times of the year (Chan et al., 2007, Newport, 1981, Rajani
and Zhan, 1996). Failure numbers have been reported to peak during winter
months in a number of countries including Canada and the United Kingdom,
while in Australia a peak has been reported during summer. The available data
on the existence of a summer peak in failure numbers is very limited. The single
publication on the seasonal variation of pipe failures in Australia by Chan et al.
(2007), indicated that the observed peak does not occur in a consistent manner,
varying in magnitude to such a degree as to be strongly evident in some years
and scarcely observable in others. No quantitative explanation of the cause
of such variation was given. Consequently little is known about the specific
causes of seasonal failures in the Australian context.
CHAPTER 1. THESIS INTRODUCTION 3
No systematic study has been undertaken on the causes of this variation

in failure rate. Hence it is not possible to predict when failures will occur at
specific times within the year. This lack of understanding limits the actions
which can be undertaken to minimise the consequences of failure and reduce
the effectiveness of management expenditure.
1.2 Aims and scope of research

The objective of the research presented in this thesis was to investigate the
causes of the seasonal variations in the observed rate of failures occurring in
buried water reticulation pipes in Melbourne, Australia. This objective was
achieved by research undertaken in three major stages.
Stage one - An investigation of historical data in order to formulate a

hypothesis about the causes of seasonal variation in pipe failure rate.
Stage two - A detailed field study of an in-service pipe, to test the hypoth-
esis formulated in Stage One.
Stage three - The development of a modelling approach to represent the
mechanism responsible for the seasonal variation in pipe failure rate.
This research focuses on the water reticulation pipe network of Melbourne,

Australia. Melbourne has a temperate climate, typically with warm to hot
summers and cool winters (without freezing conditions). This research focuses
on buried pipes used for reticulated supply of potable water to urban proper-
ties. Whilst the findings presented are focused on the water reticulation pipe
network of Melbourne it is expected that the findings can be applied to similar
reticulation networks in other cities with similar environmental conditions.
1.3 Thesis overview

The research presented in this thesis was undertaken in three major stages,
exploratory statistical analysis, field study and modelling. This research is
presented in 9 chapters and four appendices. A summary of the content of
each chapter and appendix is given below. A series of research papers have
been written based on the content of this thesis. Where all or part of a chapter
has been used to produce a paper, the sections used are noted in the relevant
chapter’s introduction.
4
Chapter 2 examines the literature on the causes of pipe failure, focusing on

observed trends in pipe failure with respect to the attributes of the pipe and
its surrounding environment.
Stage one
Chapter 3 analyses historical reticulation network data. The analy-
sis initially determines the trends in pipe failure rate without reference
to seasonal variations. The trends in pipe failure rate are determined
with respect to pipe diameter, pipe material, pipe age, surrounding soil
type and failure type. Seasonal variations in pipe failure rates are then
investigated with respect to environmental conditions using climatic pa-
rameters. The effect of environmental conditions on pipe failure is further
investigated for interactions with pipe diameter, pipe material, surround-
ing soil type and failure type. The results of this investigation are used to
develop the hypothesis that the seasonal variation in pipe failure rates in
Melbourne, Australia occurs as the result of soil shrinkage. The portion
of the analysis which focuses on the influence of climate on pipe failure
has been reported in a paper published in a peer-reviewed international
journal.
Stage two
Chapter 4 presents the detailed field study undertaken on an in-service
pipe and its surrounding environment. The field study was undertaken to
test the hypothesis proposed in Chapter 3 by systematically investigat-
ing the response of an in-service pipe subject to real world environmental
conditions. Pipe strain, soil water content, soil suction, soil movement
and weather conditions (rainfall and air temperature) were the primary
parameters recorded as part of this study. This chapter includes site-
selection, instrumentation design, installation of instrumentation and
analysis of corrected data. This chapter does not include the analysis
of the raw data collected from the field. The analysis of the corrected
data confirmed the hypothesis that seasonal variation in pipe failure rates
in Melbourne, Australia occurs as the result of soil shrinkage.
The field study detailed in this chapter was undertaken in concert with
a second PhD student. Unless otherwise noted the field work was under-
taken equally.
CHAPTER 1. THESIS INTRODUCTION 5
Stage three
The aim of Stage three of the research presented in this thesis is to develop
a model to predict the development of stresses in pipes resulting from soil
shrinkage. To do this, a number of pieces of work were undertaken; the me-
chanical properties of the pipe to be modelled must be known (Chapter 5), as
must the behaviour of the soil surrounding the pipe in response to changes in
water content (Chapters 6 and 7). The results from these investigations are
then combined to produce the final model (Chapter 8).
Chapter 5 investigates the mechanical properties of Australian cast iron

pipes. As a comprehensive investigation of the mechanical properties of
cast iron pipe in North America and less comprehensive studies from
the United Kingdom and Japan have already been undertaken, the work
detailed in this chapter was conducted as a scoping study to allow a
comparison with these data. It was found that the mechanical properties
of Australian cast iron pipes were comparable to the results reported in
the literature.
Chapter 6 presents a novel technique for creating empirical equations to

model soil behaviour. The application of this technique is demonstrated
through creation of a new model which describes the soil water charac-
teristic curve. The work presented in this chapter has been reported in
a paper submitted to a peer-reviewed international journal.
Chapter 7 applies the technique detailed in Chapter 6 to develop an

empirical equation to model the soil shrinkage curve. This model is then
extended to incorporate the effect of net stress on soil volume change
behaviour. The work presented in this chapter has been reported in a
paper published in a peer-reviewed international journal.
Chapter 8 presents the numerical model developed to represent the in-

teraction between a buried pipe and its surrounding soil. This model is
validated by using it to represent the interaction of the pipe and sur-
rounding soil instrumented during the field study presented in Chapter
4.
Chapter 9 of this thesis presents the conclusions of this thesis and provides
recommendations for future research.
6
Appendix A details the data preparation undertaken prior to the statistical

analysis presented in Chapter 3.
Appendix B contains a summary of the raw data collected from the field
instrumentation site. The data shown here was collected from the times and
dates detailed in Table 4.3 until 17th of March 2010.
Appendix C details the steps taken in the development of the equation for
the soil shrinkage curve presented in Chapter 7.
Appendix ?? contains those publications resulting from the research under-

taken as part of this degree which have not been included in the main body of
this thesis.
Chapter 2
Factors affecting buried pipe

failure
2.1 Introduction
This chapter investigates the current state of knowledge with regards to the
failure of buried water reticulation pipes, focusing on the identification of the
causes and types of buried reticulation pipe failure. Incorporated in this is the
identification of observed trends in failure with respect to physical attributes
of the pipe and its surrounding environment.
Section 2.2 defines pipe failure and describes the causes of failure. Section
2.3 investigates pipe structural capacity and the modes of capacity deterio-
ration. Section 2.4 investigates the sources of loads on buried pipe and the
types of stresses these loads induce in the pipe. Section 2.5 details the types
of failure observed in buried water reticulation pipes. Section 2.6 investigates
observed trends in pipe failure with respect to the physical attributes of the
pipe and it surrounding environment.
2.2 Defining pipe failure

The failure of a water reticulation network occurs when that network is no
longer able to perform its primary objective. The primary objective of a water
supply system is to provide water to meet the demands imposed upon it by
customers. A secondary function of a water supply system is to provide water
for fire fighting. Wakool Shire Council (2000) state that ‘customer require-
ments shall be met by providing a water main and allowing an appropriate
point of connection for each individual property.’ For failures at the network
7
8
level to occur, a large number of pipe failures would be required, an unlikely

occurrence in a maintained network.
In terms of the reticulation network failure, a pipe within that network
could be said to fail when it is no longer able to transport water between adja-
cent pipes and to customers connected to that pipe. However, this definition
is not appropriate, as a pipe is still able to perform this function whilst losing
significant amounts of water from breaches of the wall of the pipe, that oc-
cur as the result of structural failure. Alternately a pipe may not be able to
perform this function without undergoing structural failure due to hydraulic
failure (Olliff and Rolfe, 2002).
For the purposes of this thesis failure of a pipe is said to have occurred
when the structural integrity of the pipe has been lost and the pipe undergoes
mechanical failure. Other forms of pipe failure are not considered here. Me-
chanical failure occurs when the load(s) applied to the buried pipe exceeds the
structural capacity of the pipe to resist them.
The load applied to pipes do change over time and can cause failure; how-
ever in practice, the failure of a pipes is more likely result when their structural
capacity has reduced to a level where they are no longer able to resist their
loading conditions (Davis et al., 2004).
2.3 Pipe structural capacity and its deteriora-

tion
Pipe structural capacity is the measure of the ability of the pipe to withstand
loads. Pipe structural capacity is described in the literature in terms of two
measures, tensile strength and fracture toughness. The failure mechanism
when the tensile strength or fracture toughness of a material is exceeded is
referred to here as net section collapse and fracture respectively. The mode of
failure is dependent on which of these measures of pipe capacity is exceeded
and on its applicability to the specific pipe material. Atkinson et al. (2002)
found that for cast iron (CI) pipes failure can occur as a result of net section
collapse for small defects in the pipe wall (approximately up to 30% of pipe
wall thickness) above this size failure would occur as fracture. The actual
transition between failure mechanisms varies dependent on the relative values
of tensile strength and fracture toughness.
The presence of defects in a material would result in a reduction of capacity
from the ideal situation. Defects in pipe material can exist for several reasons
CHAPTER 2. FACTORS AFFECTING BURIED PIPE FAILURE 9
including mechanical damage (such as scratches and gouges), manufacturing

flaws (Burn et al., 2005, Makar et al., 2001) and environmentally induced
defects such as corrosion pits. Different to the other sources of pipe defect,
environmentally induced defects develop in the pipe over time after installation
and are the cause of capacity deterioration. The mechanism of pipe capacity
deterioration is dependent on the pipe material.
The predominant deterioration mechanism of CI pipes is electro-chemical
corrosion (Makar, 1999, Rajani and Kleiner, 2001). Where the metallic matrix
is removed and the graphite flakes retained, this process is known as graphiti-
sation. Ductile iron (DI) pipe also undergoes electro-chemical corrosion; for DI
this does not result in graphitisation. Polyvinyl chloride (PVC) pipes experi-
ence slow crack growth from mechanical damage or manufacturing flaws (Davis
et al., 2007b). An example of a manufacturing flaw in the wall of a PVC-U pipe
is shown in Figure 2.1, evidence of slow crack growth prior to failure can be
seen in the form of striations radiating from the flaw. Polyethylene (PE) pipes
also experience slow crack growth although following a significantly different
mechanism (Davis et al., 2008a). Asbestos cement (AC) pipes deteriorate due
to the loss of calcium hydroxide (CaOH) (Al-Adeeb and Matti, 1984, Davis
et al., 2008b) and as a result of sulphate attack (Matti and Al-Adeeb, 1985).
Models for the deterioration of CI, AC, PVC and PE pipe capacity have been
proposed by Davis et al. (2004, 2007a, 2008a,b) based on these mechanisms.
Figure 2.1: Manufacturing flaw in wall of PVC pipe
2.4 Sources of buried pipe loading

Buried pipes are subject to multiple loads acting simultaneously after instal-
lation. These loads have been segregated here into ordinary and non-service
loads, Sections 2.4.1 and 2.4.2 respectively. Current structural design of buried
pipes is based primarily on the stresses developed as a result of internal pres-
sure and external loads (Rajani and Tesfamariam, 2004). For the purpose of
10
this review these loads are considered here as ordinary loads. Loads acting in
addition to ordinary loads are defined here as non-service loads.
This section does not attempt to provide detail on the calculation of loads
to which a buried pipe is exposed.
2.4.1 Ordinary loads

The ordinary loads to which a buried pipe are subject are in-plane external
loading from dead and live loads, and circumferential loading due to internal
pressure (Moser, 2001), as shown in Figure 2.2.
Live load
Dead load
Internal
pressure
Figure 2.2: Schematic representation of ordinary loads acting on buried pipe
2.4.1.1 Internal pressure
Internal water pressure induces circumferential stresses in the pipe wall. The
magnitude of the circumferential stress induced is dependent on the diameter
of the pipe and it wall thickness, as shown in Equation 2.1 (Moser, 2001).
Equation 2.1 assumes that the pipe can be treated as a thin walled cylinder,
defined as when D/t < 10.
Pi (Do − 2t)
σ= (2.1)
2t
where σ is the maximum stress in the pipe wall, Pi is the internal pressure, Do
is the pipe outside diameter and t is the wall thickness.
2.4.1.2 Dead load
Dead loads are constant loads that are the result of gravity acting on the soil
and any other static components of the environment above the pipe, such as
pavement. The application of dead loads to the pipe is not simply determined
by the mass above the pipe. Instead, the loads imposed are dependent on
the relative stiffness properties of both the pipe and the soil. This results in a
statically indeterminate system as the load supported by the pipe is determined
by the deflection of the pipe relative to the soil and the deflection of the pipe
determines the load supported by the soil (Moser, 2001, Watkins and Moser,
1998).
Where the buried pipe is stiff relative to the surrounding soil, these pipes
are referred to as ’rigid’. Rigid pipes support the majority of the load in the
pipe-soil system. The dead load applied to a rigid pipe can be calculated
following Marston load theory. For a pipe buried in a trench the maximum
load applied to the pipe will be that of all soil backfilled into the trench, if the
friction between the backfilled soil and the undisturbed soil surrounding the
trench and cohesion within the backfilled soil are assumed to be negligible.
The load applied to the rigid pipe is calculated as shown in Equation 2.2.
Wt = Cd γBd2 (2.2)
where Wt is the maximum Marston load per unit length, Cd is the load co-
efficient, γ is the unit weight of the soil and Bd is the trench width. Cd is
dependent on soil type and the ratio of burial depth and Bd .
Where the buried pipe is compressible relative to the soil the pipe is con-
sidered to be ’flexible’. The load on a flexible pipe is substantially less than on
a rigid pipe under the same burial conditions (Moser, 2001). In the long term
the load supported by a flexible pipe approaches the prism load. The prism
load is the load due to the soil vertically above the pipe. The prism load is
calculated as shown in Equation 2.3.
σp = γDDo (2.3)
where σp is the prism load, γ is the unit weight of the soil, D is soil depth
above the pipe, Do is the pipe outside diameter.
12
2.4.1.3 Live load
Live loads occur as the result of objects, such as vehicles, moving over the pipe
(Watkins and Moser, 1998). Not all buried pipes are subjected to live loading,
as the exposure to this load is dependent on pipe location. For example, a
pipe located in a nature strip is unlikely to be subject to live loading on a
regular basis, although the possibility cannot be excluded. A comparable pipe
which crosses under a road will be subject to live loading on a regular basis,
the magnitude of which is dependent on the traffic on road.
As live loads are transferred to the pipe by the soil, the proportion of the
live load experienced by the pipe is dependent on pipe depth (Trickey and
Moore, 2007). The proportion of the concentrated live loads can be calcu-
lated by integration of the Boussinesq solution, as shown in Equation 2.4.
The Boussinesq solution gives the stress distribution in a semi-infinite elastic
medium due to a point load applied at its surface. This solution assumes an
elastic, homogeneous, isotropic medium. Whilst soil is not such a medium,
experiments have shown that the Boussinesq solution gives reasonably good
results for soil (Moser, 2001).
Cs P F 0
Ws d = (2.4)
L
where Ws d is the live load per unit length, Cs is the load coefficient (dependent
on burial depth and pipe diameter), P is the concentrated load, F 0 is the impact
factor and L is the effective pipe length.
2.4.2 Non-service loads
Non-service loads are defined here as those loads which act on a buried pipe
in addition to ordinary loads. Non-service loads include surge pressure, ther-
mal expansion/contraction, frost loading, loss of support and soil shrink/swell.
Whilst this list does not encompass all possible non-service loads it does cover
the most common types. Kettler and Goulter (1985) noted that for CI pipe
which have undergone graphitisation failure occurs as a result of non-service
loads. Rajani et al. (1996) reported on failure modes for water pipes in North
America, all of which were caused by non-service loads. Figure 2.3 shows a
schematic representation of common non-service loads.
Restrained thermal
expansion/contraction
Surge pressure
or Freezing of water
Hogging - soil swelling, frost load
Sagging - loss of support, soil shrinkage
Figure 2.3: Schematic representation of common non-service loads acting on

buried pipe
2.4.2.1 Soil shrinkage/swelling
Soil shrinkage and swelling occurs as a result of soil water content change. The
magnitude of soil volume change is dependent on the shrink/swell potential
of the soil and the change in the soil water content. Swelling of soil causes
the pipe to bend, imposing longitudinal tensile stresses (Rajani et al., 1996).
Conversely the shrinkage of soil due to a reduction in soil water content also
applies longitudinal loads due to bending.
In both cases, pipe bending occurs as a result of differential soil movement.

Limited data are available on the magnitude of the loads imposed on buried
pipes as a result of soil shrinkage and swelling. This lack of data was noted by
Moser (2001) and Kleiner and Rajani (2002). Moser (2001) commented that
this may be due to design practises preventing pipes being buried directly in
soils with high shrink/swell potential to prevent the negative effects of such
soils. However, the increasing use of trenchless installation techniques (Dyt,
2009) means that pipes are currently being installed directly in the native soil
without the use of imported backfill to offset the effect of soil volume change.
14
2.4.2.2 Loss of support
Loss of support and inadequate bedding of pipes results in uneven support

along the length of the buried pipe. The uneven support condition along
with the ordinary vertical loads (dead and live loads) causes pipe bending.
Pipe bending introduces flexural stresses in the pipe acting in the longitudinal
direction, similar to that seen as a result of differential soil shrinkage.
Rajani and Tesfamariam (2004) proposed an analytical model that can be
used to calculate longitudinal stress developed in a pipe as a result of lost
support. It is important to note that unlike differential soil shrinkage, a loss
of support assumes the complete removal of soil beneath portions of the pipe.
The model predicted that longitudinal stress increases as soil stiffness in-
creases, pipe diameter decreases, and pipe stiffness increased. It was seen that
whilst the stresses developed in stiffer CI pipe were higher than those devel-
oped in PVC pipe, the stress developed in the PVC pipe was significantly
higher as a proportion of ultimate strength.
2.4.2.3 Frost loading
Frost loading develops as a result of frost penetration into the soil and frost
heave, where an increase in both factors will increase the resultant loading.
Frost loading develops as a result of the expansion of in situ soil water as it
freezes, creating ice lenses. The freezing of soil water has a drying affect on
the soil causing water to move to the frost front by capillary action (Zhan
and Rajani, 1997). Smith (1976) reported experimental results showing that a
frost depth of two feet was able to increase vertical load on a pipe buried at a
depth of 4.5 feet by 80%, the maximum measured vertical loading more than
doubled due to the action of frost. Baracos et al. (2010) reported that for the
clay soils in Winnipeg, Canada the shrinkage resulting from the reduction of
soil water content was greater than the expansion caused by freezing. Rajani
and Zhan (1996) and Zhan and Rajani (1997) proposed analytical models
to predict the magnitude of the load resulting from frost loading depending
on environmental conditions. Trickey and Moore (2005) presented a three
dimensional finite element model for the effect of differential frost heave on a
CI pipe.
2.4.2.4 Surge pressure
Surge pressure results from a pressure differential due to changing flow velocity,
as happens on the closing or opening of valves, producing a shock wave which
propagates along the water column (Webb and Gould, 1978). Surge pressure,
also know as transient surge, is often referred to as water hammer (Moser,
2001).
The magnitude of pressure experienced during a surge pressure event is
dependent on the geometry of the system, the magnitude of water velocity
change and the velocity of the pressure wave as shown in Equation 2.5 (Moser,
2001).
a
∆P = ∆V (2.5)
g
where ∆P is the surge pressure, a is the velocity of the pressure wave, g is
gravitational acceleration and ∆V change in flow velocity.
The occurrence of surge pressure is dependent on the time over which the
velocity change happens. By reducing the rate at which the flow velocity
changes water hammer can be minimised or avoided (Webb and Gould, 1978).
2.4.2.5 Restrained thermal contraction
Thermal contraction results in the development of longitudinal tensile stresses

when contraction is prevented as a result of pipe restraint. The level of stress
developed is dependent on the pipe materials coefficient of thermal expansion,
the friction between the pipe and the surrounding soil, axial restraint within
joints and the temperature change. Rajani et al. (1996) presented a model
to determine longitudinal stress developed in buried pipe as a result of pipe
temperature change, the pipe-soil reaction modulus and internal pressure. The
model accounted for friction between the pipe and the surrounding soil via the
pipe-soil reaction modulus.
2.5 Types of pipe failure

The type of pipe failure which occurs is dictated by the loading conditions to
which that pipe is subject. The two most prevalent failure types seen in buried
pipelines are longitudinal failures (also known as long splits) and circumfer-
ential failures (also known as circle breaks, broken backs or broken bellies).
Rajani et al. (1996) reported that on average 70% of pipe failures in water
reticulation networks are circumferential failures and 30% are longitudinal or
16
other failure types. The proportion of circumferential failures increases to 80%

for small diameter pipes (< 200 mm) (Makar et al., 2002). Contrary to this,
Goulter and Kazemi (1989) reported for the city of Winnipeg, Canada that
hole, joint, circumferential and longitudinal failures accounted for 38%, 30%,
18% and 6% of failures respectively. Newport (1981) reported that circumfer-
ential failures accounted for well over half of the observed failures.
2.5.1 Circumferential failure

Circumferential failure result from longitudinal tensile stresses induced in the
pipe by axial loading and/or flexure (Makar et al., 2001, Rajani et al., 1996).
Such stresses can result from restrained thermal contraction and pipe bending.
Restrained thermal contraction was identified by Rajani et al. (1996) as a cause
of failure in Canadian pipe reticulation networks. Pipe bending as a result
of inadequate bedding and soil swelling can impose additional longitudinal
stress in pipes (Rajani et al., 1996). Rajani and Tesfamariam (2004) and
Olliff and Rolfe (2002) reported that water leaking from a pipe can, depending
on soil composition and water flow rate, wash away the soil supporting the
pipe resulting in circumferential failure. Figure 2.4(a) shows a photographic
example of circumferential failure in a CI pipe. Figures 2.4(b) and 2.4(c)
show schematic representations of circumferential failure by axial and flexural
loading respectively; the arrows shown indicate pipe stress direction.
2.5.2 Longitudinal failure

Longitudinal failure occurs as a result of circumferential tensile stresses in
the pipe wall (Rajani et al., 1996). Such stresses can result from a number
of different load sources. Internal pressure and external loads (soil cover, live
loads and frost loads) are possible causes of longitudinal fracture (Makar et al.,
2001, Hu and Hubble, 2007). In addition to internal pressure, soil cover, live
loads and frost loads, circumferential stresses can also be induced by surge
pressure and freezing of the water inside the pipe. Makar et al. (2001) reported
that longitudinal failures in CI pipes appeared to occur exclusively on larger
diameter pipes.
The combined action of internal pressure and external loads have been used
by Olliff and Rolfe (2002) in a model to determine the factor of safety for a given
buried pipe. Davis et al. (2009) combined his model for the reducing capacity
of CI pipe with the model from Olliff and Rolfe (2002) to predict remaining
pipe lifetime. This model is only applicable to longitudinal failure types as

it only incorporates ordinary loads which induce circumferential stresses in a
buried pipe.
Figure 2.5(a) shows a photographic example of longitudinal failure in CI
pipe. Figure 2.5(b) shows a schematic representation of a longitudinal failure,
the arrows shown indicate pipe stress direction.
2.5.3 Split bell
Split bell is caused by differing coefficients of thermal expansion between the

pipe material and the material used to form a seal at the joints. Stress develops
at the joint in very cold temperatures inducing circumferential stress to develop
at the pipe bell. These stresses are limited to the area of and adjacent to the
pipe bell. It is for this reason that this failure type is separate from the
longitudinal failure type (Makar et al., 2001). This type of failure was not
identified as commonly occurring in Australia (Water Services Association of
Australia, 2003).
2.5.4 Blown section
Blown sections (also known as blow outs, blowout holes or piece blown out)
occur when the capacity of the pipe has reduced over an area and this area is
’blown’ out of the pipe wall by internal pressure (Makar et al., 2001). Rajani
et al. (1996) stated that this type of failure occurred as the result of unusually
high surge pressures. A photo of a blown section failure in a CI pipe is shown
in Figure 2.6(a).
2.5.5 Through-wall defects
Through-wall failure occurs when a defect penetrates the full wall thickness of
a pipe so that water can escape. This type of failure, can induce mechanical
failure, either by acting as a failure initiation location or by allowing sup-
port beneath the pipe to be washed away increasing pipe loading (Olliff and
Rolfe, 2002, Rajani and Tesfamariam, 2004). Figure 2.6(b) shows a photo of
a through wall corrosion pit in a CI pipe. Through wall pipe defects are also
known as perforations (Water Services Association of Australia, 2003)
18
2.5.6 Third-party damage

Third-party damage occurs when a party unrelated to the pipe network damage
the pipe, such as when a roadside construction crew breaks the pipe during
excavation works. Figure 2.6(c) shows a photo of a third-party failure which
occurred as a result of an excavator damaging the wall of a PVC pipe.
2.5.7 Tapping failure

Pipe tapping is a common action where a hole is drilled into the pipe wall
for operational reasons, such as the installation of a service connection. Tap-
ping operations can introduce defects into the pipe wall that result in pipe
failure. Figure 2.6(d) shows an example of a tapping failure which occurred in
a PVC pipe. Tapping failures were classified by Water Services Association of
Australia (2003) as a separate failure type but have been included as part of
third-party failure types as this failure type occurs as the result of a human
interference with the pipe.
(a) Photo of circumferentially failed CI pipe
(b) Schematic representation of axial loading
(c) Schematic representation of flexural loading
Figure 2.4: Circumferential pipe failure

20
(a) Photo of longitudinally failed CI pipe (failure includes blown section)
(b) Schematic representation
Figure 2.5: Longitudinal pipe failure
(a) CI pipe with blown section failure (b) CI pipe with through wall corrosion pit
(c) PVC-U failure resulting from third party (d) PVC-U pipe which failed due to a defect
damage introduced during tapping
Figure 2.6: Examples of other failure types

2.6 Factors affecting pipe failure rate

The failure of pipes is affected by a number of factors related to the pipe and
its environment. Kleiner and Rajani (2002) reported that pipe failure rates
were affected by the numerous characteristics of the pipe and its environment.
A summary of the static and dynamic factors is shown in Table 2.1. This
section describes trends reported in the literature on how pipe diameter, pipe
material, soil properties, temperature, age and season affect pipe failure rates.
The remaining factors have not been included due to limited data on the trends
in observed failure rate related to changes in these attributes.
Table 2.1: Factors affecting pipe failure rate (Kleiner and Rajani, 2002)
Static Dynamic
Pipe material Pipe age
Pipe diameter Temperature (soil, water)
Wall thickness Soil water content
Soil (backfill) characteristics Soil electrical resistivity
Installation Bedding condition
Dynamic loading
2.6.1 Pipe material

Pipe material has a significant affect on observed pipe failure rates. As de-
scribed in Section 2.3, the mechanism of pipe capacity reduction is dependent
on pipe material. Additionally, pipe material affects how the pipe responds to
applied loads. As pipe materials vary with respect to a number of properties,
this section compares pipe materials by failure rate.
The failure rates reported for specific materials vary widely between loca-
tions. Where failure rates are reported, often values are only given for a single
material. This section will report on the relative ranking of pipe materials by
failure rate, where the failure rates for multiple materials were given. The re-
ported failure rates for single materials have not been included as they cannot
be used for a reliable indication of relative failure rate between locations.
Vloerbergh and Blokker (2007) reported failure rates for AC, CI, DI and
PVC/PE pipes. Combined failure rates were given for PVC and PE pipe.
CI had the highest failure rate, followed by AC. PVC/PE had approximately
half the failure rate of AC, followed by DI with the lowest failure rate. In
contrast to these results for DI and PVC/PE, Pelletier et al. (2003) reported
significantly higher failure rates for DI than for PVC.
22
Pipe materials are also categorised as either flexible or rigid. The categori-
sation of pipe material as either flexible or rigid is based on the compressibility
of the pipe relative to the surrounding soil. A pipe is considered as rigid when
the soil is compressible relative to the pipe. Conversely, when the pipe is com-
pressible relative to the soil the pipe is considered to be flexible (Watkins and
Moser, 1998).
The underlying reason for the material being ranked (with respect to failure
rate) in the order shown is not clearly discussed in the literature. However,
it seems that the underlying reason is related to the age and flexibility of the
different pipe materials. Moser (2001) categorised CI and AC as rigid, and
DI, PVC and PE as flexible. PE would be expected to have the lowest failure
rates as it is the most recent pipe material to be introduced (Scott, 1990).
Davis et al. (2007a) reported that the original PE used to manufacture pipes
exhibited much higher than expected failure rates and since then the quality
of the PE material used has improved significantly. Additionally Davis et al.
(2007a) reported that joint failures in PE pipe are common.
2.6.2 Pipe diameter

Statistical data relating pipe failure rates to pipe diameter have been reported
by numerous authors (Boxall et al., 2007, Gorji-Bandpy and Shateri, 2008, Hu
and Hubble, 2007, Hudak et al., 1998, Kettler and Goulter, 1985, Newport,
1981, Pelletier et al., 2003, Rajani and Zhan, 1996, Vloerbergh and Blokker,
2007). All authors reported that data showed a trend of increasing failure
rate as pipe diameter decreases. This trend was observed where the data were
limited to pipes with a minimum pipe diameter of DN 100 mm. Where pipe
diameters are reported with the prefix DN these values are nominal only and
do not refer to actual values, e.g a DN 100 mm pipe has an actual diameter
of 121.9 mm. The data reported by Gorji-Bandpy and Shateri (2008) and
Vloerbergh and Blokker (2007) for pipes <DN 100 mm showed that failure
rate decreased below DN 100 mm.
Kettler and Goulter (1985), Hu and Hubble (2007) and Rajani et al. (1996)
proposed differing causes for the observed trend. Each theory primarily re-
lates to differing failure scenarios. Kettler and Goulter (1985) proposed that
the observed trend was a result of larger diameter pipes having thicker walls
and more reliable joints. The increase in wall thickness with diameter would
result in pipe with higher structural capacity and therefore be less likely to
fail structurally as a result of external loads. Additionally, if corrosion was the
primary cause of capacity reduction the increased wall thickness would reduce
the observed failure rate.
Hu and Hubble (2007) proposed that the decrease in failure rate as pipe
diameter increased was the result of increasing moment of inertia, reducing
the stress which develops in the pipe as a result of bending.
Rajani et al. (1996) proposed a model for axial pipe stress development due
to seasonal temperature extremes. This model predicted that as pipe diame-
ter decreases the maximum axial tensile stress developed increases (maximum
and minimum temperatures used in modelling were 30 °C and 0 °C respec-
tively). This result was obtained for DI pipes between DN 200 mm and DN
800 mm. This trend was not followed for the DN 100 mm pipe modelled,
where maximum axial tensile stress developed was seen to decrease. For each
pipe modelled, wall thickness increased with diameter, with the exception of
the DN 100 mm pipe which had the same wall thickness as the DN 200 mm
pipe.
2.6.3 Surrounding soil

Clark (1971), Goulter and Kazemi (1988), Hu and Hubble (2007), Hudak et al.
(1998), Kleiner and Rajani (2002), Newport (1981) and Rajani et al. (1996)
reported that the type of soil surrounding the pipe can affect the pipe fail-
ure rate. Specifically these authors reported that soils with high shrink/swell
potential result in higher pipe failure rates. In contrast, Boxall et al. (2007)
reported that the shrink/swell potential of the soil had no significant affect on
pipe failure rate.
Hudak et al. (1998) reported that pipe failure rates increased as soil plas-
ticity increased (soil plasticity is an indication of shrink/swell potential). A
significant number of failures was believed to occur as result of rain wetting
dry soil, failure also occured due to soil shrinkage during dry months (Hudak
et al., 1998). Hu and Hubble (2007) also reported that pipe failure occurred as
the result of pipe bending due to soil shrinkage in high shrink/swell potential
soil.
Clark (1971) reported anecdotal evidence which related soil with a high
shrink/swell potential to the failure of buried pipes due to bending. However,
soil swelling was identified as the cause of bending. A study by Kassiff and
Zeitlen (1962) on the stresses induced in AC pipes buried in a soil with high
shrink/swell potential as a result of seasonal changes in soil water content
and irrigation had similar results. Differential swelling in both vertical and
24
lateral directions was found to result in high stresses, several times greater
than those caused by internal pressure, developing in the pipes tested. Clark
(1971) reported that rigid pipes are more susceptible to damage from soil
swelling as flexible pipes are able to bend to allow soil expansion to occur,
relieving the applied load.
2.6.4 Pipe age

The literature reports conflicting data on the relationship between pipe age and
failure rate. Newport (1981) reported that pipe age is not a good indicator of
condition, and therefore failure. Whilst not clearly stated it is believed that
this study focused on CI pipes. Vloerbergh and Blokker (2007) also found no
relationship was found for CI, PVC, PE or steel pipes. In contrast, Kettler
and Goulter (1985) reported an increase in CI failure rates with pipe age,
primarily associated with joint failures, and Boxall et al. (2007) reported that
CI failure rate initially decreased with age until approximately 80 years before
then increasing.
For AC the trend is clearer, Boxall et al. (2007), Hu and Hubble (2007),
Kettler and Goulter (1985) and Vloerbergh and Blokker (2007) all reported
an increase in failure rates increased with age.
Pipe age is also commonly used in statistical models for the prediction
of pipe failure, where the probability of pipe failure increased with age for
any single pipe type (Constantine and Darroch, 1993, Constantine et al., 1996,
Jarrett et al., 2001, 2002, 2003, Kleiner and Rajani, 2001, 2002, Pelletier et al.,
2003, Shamir and Howard, 1979, Walski and Pelliccia, 1982).
The deterioration of pipe with age is known to occur, however the relation-
ship between pipe age and failure is less clear due to the change in environ-
mental conditions which control the rate of deterioration across pipe networks.
Therefore, pipe age alone is not a reliable indicator of pipe condition.
2.6.5 Season
Several authors have reported that pipe failure rates peak at different times of
the year. These peaks were commonly related to conditions associated with
the different seasons. Generally speaking, the reported peaks were observed
either in winter (Ciottoni, 1985, Goulter and Kazemi, 1989, Habibian, 1994,
Rajani and Zhan, 1996), summer (Chan et al., 2007, Hudak et al., 1998) or
in both (Baracos et al., 2010, Hu and Hubble, 2007, Newport, 1981). Baracos
et al. (2010) and Hu and Hubble (2007) reported major peaks in summer and
minor peaks seen in winter. In contrast Newport (1981) reported major peaks
in winter and minor peaks in summer.
2.6.5.1 Winter
The literature reported in this section presents data for pipe networks located
in Canada, the UK and Northern USA. These regions all experience freezing
conditions during winter. No peaks in pipe failure rate were reported for
regions, such as Australia, where freezing conditions are not experienced.
Ciottoni (1985) reported that in New York, USA approximately 45% of
pipe failures occurred during the three coldest months of the year. Habibian
(1994) reported a direct effect of temperature in Washington D.C., USA. Low
water temperatures were observed to result in an increase in pipe failures, with
the change in water temperature being a significant factor affecting the number
of failures. The majority of failures associated with low water temperatures
were circumferential in type and occurred in small diameter CI pipes.
Rajani et al. (1996) proposed an analytical pipe-soil interaction model to
predict longitudinal stress developed in a pipe as a result of temperature and
internal pressure. This model showed that longitudinal stress increased as
ground temperature decreased as a result of restrained thermal contraction.
The magnitude of the stresses developed increased as the temperature change
between installation (which is undertaken during summer) and minimum win-
ter pipe temperature increased.
Frost loading has also been associated with the winter peak in pipe failure
rates. Newport (1981) reported that in the UK the observed increase in pipe
failure rates during winter was directly related to frost loading, based on the
strong correlation observed between the freezing index and the cumulative
number of pipe failures observed. The freezing index is the cumulative average
daily temperature below 0 °C within a given time period. Walski and Pelliccia
(1982) statistically modelled pipe failures for Binghamton, NY, USA using
the temperature in the coldest month as an indicator of frost penetration to
account for some of the inter-year variation in failure numbers. Rajani and
Zhan (1996) also observed a strong correlation between freezing index and the
observed number of pipe failures for North York, Canada.
Hu and Hubble (2007) and Kleiner and Rajani (2002) conducted statistical
analyses on historical pipe failures for Regina and Ottawa, Canada respectively.
Kleiner and Rajani (2002) did not explicitly analyse failures on a seasonal ba-
26
sis, but did include parameters in the analysis which were related to specific
seasons, i.e., freezing index and snapshot rainfall deficit. Hu and Hubble (2007)
also reported a strong relationship between the freezing index and pipe failure
when only winter failures were analysed (as noted above Hu and Hubble (2007)
reported a minor peak in pipe failures during winter). When all pipe failures
were analysed (winter and summer failures combined) the freezing index was
not found to have a strong affect on pipe failure. The authors believed this
to be a consequence of the high number of summer failures (almost double
the number of winter failures) and that frost penetration is an important con-
tributing factor to winter pipe failure. Kleiner and Rajani (2002) reported
that the freezing index was not statistically significant in pipe failure. This re-
sult was contrary to what they had expected and it is believed that the burial
depth of pipes in Ottawa was deep enough to sufficiently reduce the effect of
frost loading.
Hu and Hubble (2007) and Kleiner and Rajani (2002) both reported on
the influence of rainfall deficit on annual failure rates. Rainfall deficit was
used as a surrogate indicator of soil water content, where a high rainfall deficit
indicates low soil water content. Rainfall deficit had a strong correlation with
failure rate. Kleiner and Rajani (2002) assumed that a high rainfall deficit
would result in greater frost penetration. The rainfall deficit at the start of
winter was used for this purpose, known as the snapshot rainfall deficit. The
average annual rainfall deficit was used as a measure of soil dryness and is
discussed in Section 2.6.5.2.
Hu and Hubble (2007) found snapshot rainfall deficit had a strong negative
relationship with pipe failure. Hu and Hubble (2007) indicated that the higher
soil water content (indicated by the negative relationship with rainfall deficit)
coupled with the freezing index may have contributed to the winter peak in
pipe failures. Kleiner and Rajani (2002) did not find the snapshot rainfall
deficit to be statistically significant, as was their belief with the freezing index
it is believed that the low significance was due to the high burial depth of the
pipes in Ottawa.
Baracos et al. (2010) suggested that winter pipe failure was correlated
with dried soil conditions occurring just prior to the spring thaw based on
observations from Winnipeg, Canada.
The literature presented above indicates that increases in pipe failure rate
observed during winter were related to low soil and water temperatures, frost
loading and soil water content.
2.6.5.2 Summer
The literature reported in this section presents data for pipe networks located
in Canada, the UK, USA and southern Australia. The increase in the number
of pipe failures occurring during summer has received less attention than those
occurring during winter.
An increase in the number of pipe failures during summer was observed for
Victoria, Australia and Texas, USA by Chan et al. (2007) and Hudak et al.
(1998) respectively. Hudak et al. (1998) reported that the shrinkage of soil due
to summer conditions contributed to an increase in the number of pipe failures
at that time. Chan et al. (2007) also observed an increase in the number of
pipe failure numbers during summer due to soil shrinkage. Newport (1981)
observed a slight increase in number of pipe failures occurring during summer
in the UK. This slight increase was believed to be the result of extreme drying
conditions causing an increase in ground forces.
The importance of soil shrink/swell potential was highlighted by Chan
et al. (2007) who segregated pipes within a single reticulation network as be-
ing buried in either soil with high shrink/swell potential or soils with low
shrink/swell potential. The authors found that pipes buried in soils with high
shrink/swell potential exhibited approximately twice the rate of failures on
average over an eight year period compared to pipes buried is soils with low
shrink/swell potential. The failure rates observed in two separate summer
months within that eight year period for pipes in soils with high shrink/swell
potential was approximately five times that of pipes buried in soils with low
shrink/swell potential.
Hu and Hubble (2007) and Kleiner and Rajani (2002) both reported on
the influence of rainfall deficit on annual failure rates in statistical models.
Hu and Hubble (2007) found that rainfall deficit had a very strong positive
relationship with AC pipe failure in the Regina region, which they associated
with the high shrink/swell potential of the local soil. Kleiner and Rajani (2002)
reported a similar result for Ottawa, Canada where Leda clay (a clay with a
high shrink/swell potential) comprises a significant amount of the soil.
The literature presented indicates that increases in pipe failure rate ob-
served during summer were related to soil shrinkage. However this conclusion
was primarily based upon inferred knowledge, i.e. as failures occur during sum-
mer, they are most likely to relate to a reduction of soil water content. The re-
lationship between pipe failure and summer has therefore not been rigourously
investigated to test this conclusion.
28
Rajani and Kleiner (2001) presented a comprehensive paper detailing the

physically based models developed for the structural deterioration and loading
of buried water mains. The authors explicitly noted the lack of a model to
predict the stresses which develop in buried pipes as the result of soil shrinkage.
2.7 Conclusions
This chapter has reviewed the literature to determine the current state of the
knowledge with regards to the failure of buried water reticulation pipes. The
types and causes of pipe failure have been described. The observed trends in
pipe failure with respect to physical attributes of the pipe and its surrounding
environment have been identified. These factors are pipe diameter, pipe ma-
terial, the soil type surrounding the buried pipe, the age of the pipe and the
climatic conditions prevalent at different times of the year.
The effect of the seasonal climatic conditions was identified as playing a
significant role in pipe failure, causing seasonal variations in pipe failure num-
bers. Existing research was identified in regards to the mechanism by which
seasonal climatic conditions affect pipe failure in the context of freezing en-
vironments. However, research with regards to the mechanism by which this
occurs within the context on Australian conditions has received only limited
attention. A systematic study to identify the causes of the seasonal variations
in Australian pipe failure numbers has not been undertaken.
Chapter 3
Pipe network historical failure

analysis
3.1 Introduction
This chapter comprises Stage one of the research presented in this thesis. Chap-
ter 2 identified that a systematic study to identify the causes of the seasonal
variations in Australian pipe failure numbers has not been undertaken. Such a
study is presented here using historical data for two adjacent water reticulation
networks. This analysis will determine the trends in pipe failure with respect
to pipe material, pipe diameter, surrounding soil type, pipe failure type and
pipe age. The results of this analysis are then used as the foundation for the
investigation of the effect of seasonal climatic conditions on pipe failure.
The investigation into the effect of seasonal climatic conditions on pipe
failure is undertaken in two parts. Part one identifies the correlation between
thirteen climatic parameters and monthly pipe failure rate. Part two of the
investigation examines the data for interaction between climatic conditions
and pipe material, pipe diameter, surrounding soil type and pipe failure type.
The results of this work are then used to formulate a hypothesis as to the
cause of the seasonal variation in pipe failure rate in Melbourne, Australia.
Section 3.2 presents the data supplied for this analysis. Section 3.3 details
the results of data cleaning undertaken to prepare the supplied data for sta-
tistical investigation. Section 3.4 presents the exploratory statistical analysis.
Section 3.5 details the investigation into the effect of seasonal climatic condi-
tions on pipe failure. The research presented in Section 3.5 has been presented
at an international peer-reviewed conference and has been reported in a paper
published in a peer-reviewed international journal (see List of publications and
29
30
Appendix ??).
3.2 Network data

The data used for this investigation was provided by City West Water Ltd.
(CWW) and South East Water Ltd. (SEWL), two of the three retail water
authorities operating within the Melbourne metropolitan area. The supply
areas for CWW and SEWL can be seen in Figure 3.1. Data that describes
the in-service buried reticulation pipes, henceforth referred to as assets, were
provided in GIS format. The separate GIS maps were merged into a single
dataset, ensuring that no unique identifiers were affected, and were exported
to a MS Access© database for data preparation. More details on this manip-
ulation are given below, and a summary of the data provided can be seen in
Table 3.1.
Data that describes the burst, leakage or other maintenance of buried retic-
ulation pipes, henceforth referred to as failures, were supplied by CWW in a
number of text files. These files contained all repair events recorded by CWW,
including records of work which was considered outside the scope of this inves-
tigation. From this data, the events related to pipe failures ’burst and leaks’
were isolated for use in this investigation. Failure data from SEWL were pro-
vided as a MS Excel© spreadsheet. Both datasets were imported into a MS
Access© database for separate data preparation processing prior to merging.
A summary of the failure data provided is given in Table 3.2.
Table 3.1: CWW and SEWL supplied asset data summary

Earliest Most Recent
Number Length of
Authority Construction Construction
of Assets Assets (km)
Date Date
CWW 80,744 4,201 1/01/1840 09/03/2007
SEWL 142,471 8,500 1/01/1860 26/03/2007
Total 223,215 12,701 1/01/1840 26/03/2007
Table 3.2: CWW and SEWL supplied failure data summary

Number of Earliest Failure Most Recent
Authority
Failures Date Failure Date
CWW 36,634 06/02/1979 08/05/2007
SEWL 20,174 09/07/1996 20/03/2007
Total 56,808 06/02/1979 08/05/2007
CHAPTER 3. PIPE NETWORK HISTORICAL FAILURE ANALYSIS 31
N CWW and SEWL Asset

CWW Boundary
SEWL Boundary
Figure 3.1: CWW and SEWL Assets and Supply Boundaries
3.3 Data preparation
In order for the results of this investigation to be considered reliable the data
used must first be scrutinised for reliability and unreliable records removed
(data cleaning). This process was completed in two stages, asset data cleaning
and failure data cleaning. The descriptive data in records were then grouped
by combining similar values to reduce the number of possible results (data
grouping). The details of data cleaning and grouping are not given here,
details of these processes are provided in Appendix A.
After exclusion of unreliable records during data cleaning approximately
97% of assets and 89% of failure records were retained for this investigation.
A summary of this data is given in Table 3.3 and Table 3.4.
It can be noticed in Table 3.3 that the most recent construction and failure
dates for CWW and SEWL are not the same; in fact they are significantly
different. The last date where asset and failure data is available for both
authorities is 20/09/2006. To simplify the analysis it was decided to discard
data for assets with construction dates and failures with failure dates occurring
after 31/08/2006. This excluded 831 assets (CWW 62; SEWL 769) and 3842
32
Table 3.3: Updated asset data summary

Number Length of
Date Date
CWW 77,856 4,092.0 01/01/1840 20/09/2006
SEWL 135,870 7,953.5 01/01/1860 26/03/2007
Total 213,726 12,045.5 01/01/1840 26/03/2007
Table 3.4: Updated failure data summary

Authority
CWW 28,539 06/02/1979 08/05/2007
SEWL 15,137 09/07/1996 20/03/2007
Total 43,676 06/02/1979 08/05/2007
failures (CWW 2425; SEWL 1417) from the investigation. In addition the
range of failure dates were further trimmed to create a consistent observation
period for failure data from both authorities. The start date for the observation
period was set as 01/09/1996. This allowed the data to be analysed over a
10 year period whilst retaining the maximum amount of data. This action
excluded a further 147 failures (CWW 73; SEWL 74) from the investigation.
The difference in the earliest construction date between authorities was
also investigated to determine if this was the result of missing data or due to
the east biased historical development of Melbourne city and surrounds (Vic-
toria Department of Sustainability and Environment, 2006). It was found that
whilst the oldest assets were recorded as being installed in 1840 no assets were
installed after this year until 1856. From 1856 assets were installed in each
subsequent year until 2006. For this reason the 4 assets (CWW 4; SEWL
0) installed in 1840 were removed; no failures were associated with these as-
sets. This has the effect of limiting this investigation to assets installed from
01/01/1856, Figure 3.2 shows the length of assets installed in each year from
this date. Subsequent to these actions approximately 95% of assets and 70%
of failure records supplied were retained for this investigation. A summary of
the data used in this investigation is shown in Table 3.5 and Table 3.6.
Length of assets installed (km)
Construction year
Figure 3.2: Length of assets installed in each year
Table 3.5: Asset data to analysis summary

Number Length of
Date Date
CWW 77,790 4,089.5 06/04/1856 31/08/2006
SEWL 135,101 7,917.0 01/01/1860 31/08/2006
Total 212,891 12,006.5 06/04/1856 31/08/2006
Table 3.6: Failure data to analysis summary

Authority
CWW 26,041 01/09/1996 31/08/2006
SEWL 13,646 01/09/1996 31/08/2006
Total 39,687 01/09/1996 31/08/2006
34
3.4 Exploratory statistical investigation

This section details the exploratory statistical analysis undertaken to identify
the relationships between the attributes of the pipe network based on the
literature presented in Chapter 2. This investigation will focus on the effect
of different pipe and failure attributes on the failure rates exhibited by the
combined CWW/SEWL pipe network. The failure rates described in this
section are the average failure rates over the observation period.
3.4.1 Calculation of failure rates

3.4.1.1 Recording rates
In order to compare every year of the observation period from the same basis,
i.e., number of failures occurring over a 12 month period, the recording rate
for each year was calculated. The recording rate is the proportion of the year
in which failures were recorded, recording rate was calculated on a monthly
basis. The adjustment of failure numbers using recording rates creates realistic
failure rates for years where failures were recorded for only part of the year,
allowing comparison with years with full failure records. This assumes that
those months in which failure data are available are representative of the full
year. Failure numbers are adjusted as shown in Equation 3.1. The recording
rates used in this investigation are shown in Table 3.7.
1
Adjusted Failure Number = Failure Number × (3.1)
Recording Rate
3.4.1.2 Calculation of the exposed length
The length of assets in-service changes during the observation period as the
networks under investigation are growing, as can be seen in Figure 3.2. For
this reason the length of assets exposed to failure is not constant over the
observation period. This change was accounted for in this investigation by cal-
culating the exposure of the assets during the observation period, 01/09/1996
- 31/08/2006. The exposure of assets is the time the assets were in-service
during the observation period. The age of the asset was not used as this would
cause an underestimation of failure rate as assets were in-service prior to the
observation period; however no failures were reported against them during this
time.
Table 3.7: Recording rates

Months with
Year Recording Rate
Recorded Failures
1
1996 4 3
1997 12 1
1998 12 1
1999 12 1
2000 12 1
2001 12 1
2002 12 1
2003 12 1
2004 12 1
2005 12 1
2
2006 8 3
3.4.1.3 Calculation of failure rates
Failure rates can be calculated over different time periods. For the purposes of
this investigation failure rates were calculated on a yearly basis unless otherwise
stated. Failure rate is calculated as the number of failures in each time period
recorded against the assets of interest divided by the exposed length of these
assets in that time period, Equation 3.2.
Number of Failures
Failure Rate = × 100 (3.2)
Exposed Length
3.4.1.3.1 Yearly failure rates The calculation of exposed length allows

the overall failure rates for each authority and a combined figure to be calcu-
lated, Table 3.8. It can clearly be seen that CWW suffers from a significantly
higher failure rate when compared to SEWL, Figure 3.3.
3.4.2 Pipe material

The average failure rates for each pipe material over the observation period
is shown in Figure 3.4. Examination of Figure 3.4 shows that cast iron (CI),
galvanised wrought iron (GWI), asbestos cement (AC), reinforced concrete
(RC) and oriented PVC (PVC-O) pipes had the highest failure rates by far.
The lowest failure rate of these five materials being almost double the next
highest failure rate, that of copper (CU).
Section 2.6.1 reported the relative rank of pipe materials by failure rate,
as available in the literature. Pipes made from CI and AC had the highest
36
Table 3.8: Average failure rates over observation period

CWW SEWL Combined
(failures/100 (failures/100 (failures/100
km/year) km/year) km/year)
Observation Period 70.71 18.59 35.96
†
1996 5747 12.13 26.83
1997 133.80 18.24 55.86
1998 90.13 13.34 38.52
1999 55.82 14.37 28.13
2000 61.75 18.21 32.69
2001 70.89 20.79 37.43
2002 53.41 17.53 29.47
2003 89.99 24.66 46.55
2004 70.85 22.59 38.92
2005 45.02 18.61 27.60
2006† 59.24 21.85 34.58
†
Adjusted failure rates.
failure rates, followed by the ductile iron (DI), polyethylene (PE) and unplas-
ticised polyvinyl chloride (PVC-U). The relative rank of DI, PE and PVC was
unclear due to inconsistencies between sources. The results reported here are
in agreement with the literature data for the relative rankings of CI and AC.
For the networks analysed here PE had the next highest failure rate, followed
closely by DI, with PVC having the lowest failure rate of these five materials.
As mentioned in Section 2.6.1 higher failure rates are expected to occur in

older pipe materials. This is in line with the high failure rates observed for CI,
AC, GWI, and RC as these are the oldest materials in the combined network.
However, the high failure rate of PVC-O is not expected. PVC-O is a relatively
new material (first installed in 2001 by CWW, there was no record of PVC-
O being used by SEWL) and as such was not expected to show high failure
rates. Further investigation found that PVC-O had a very short in-service
length (∼2.3 km). The short length causes the failure rate for PVC-O to be
very sensitive to any change in the number of failures. For every additional
failure, within the observation period, the failure rate of PVC-O increases by
around 4 failures/100 km/year. Therefore, the unexpectedly high failure rate
was calculated for a small number of failures, potentially associated with poor
installation practices or poor material quality.
140
Combined
CWW
SEWL
120
Failure rate (failures/100 km/year) 100
80
60
40
20
1996† 1998 2000 2002 2004 2006 †

Year
Figure 3.3: Annual failure rates over the observation period († Adjusted failure
rate)
3.4.3 Pipe diameter

The average failure rate for each diameter groups in-service over the observa-
tion period is shown in Figure 3.5. It can bee seen that from diameter group
0-50 to 100-150 failure rates increase and above 100-150 failure rates decrease
(diameter group x-y refers to those pipes with nominal diameters ≥ x and < y,
see Appendix A). Section 2.6.2 reports that failure rate is expected to increase
as pipe diameter decreases for pipe diameters above a nominal diameter of 100
mm, below this diameter failure rates are expected to decrease. This trend is
consistent with the results reported in Section 2.6.2.
3.4.4 Pipe material - pipe diameter interaction

To investigate the possibility that the trend observed in the failure rates of the
diameter groups is related to different material representation in each diameter
group, the interaction between pipe material and pipe diameter was examined.
The distribution of assets (by length) in each material/diameter group is given
numerically in Table 3.4.4 and shown graphically in Figure 3.6. It can be seen
that the group with the greatest length of assets is CI/100-150, followed by
PVC-U/100-150 and AC/100-150. It can also be seen that few materials have
38
60
50
Failure rate (failures/100 km/year)
40
30
20
10
0
AC CU GRP PE PVC-O RC
CI DI GWI PVC-M PVC-U S
Pipe material
Figure 3.4: Average failure rate by materials over observation period
more than 100 km of in-service pipe in diameter groups below 100-150. Due
to the lack of pipe length in materials other than CU, the 0-50 diameter group
has been excluded from the analysis undertaken in this section(other materials
had less than 32 km of pipe in-service when combined, over 55% of this was
PE).
Analysis of the failure rates for each material/diameter group, Figure 3.7,
shows that the failure rates for the the majority of materials (AC, CI, DI,
RC, GWI, PVC-U and S) increased with decreasing diameter. The remaining
materials (CU, PVC-M and PE) do not seem to follow this trend. However,
this observation cannot be considered as reliable due to low failure numbers,
small in-service length or a combination of both for these materials in most
diameter groups.
The length of pipe in-service and the number of failures for diameters
groups 100-150 and larger is dominated by CI and AC (these material ac-
count for around 47% of the in-service length and around 69% of the failures).
The 50-100 diameter group is dominated by PE, accounting for 70% of the
in-service length but only 18% of the failures. The high representation of PE
in the 50-100 diameter group and its low failure rate explains the low failure
Table 3.9: Length (km) of assets in-service, shown by pipe material and pipe diameter.
The material/diameter groups with the greatest length are highlighted
XX extra line extra line extra line
XXX Diameter
XXX 0-50 50-100 100-150 150-200 200-250 250-300 300-350 350-400
Material XXX
X
AC 7.47 65.29 1154.16 412.22 195.73 14.39 94.14 43.83
CI 0.30 97.41 2479.93 951.03 197.03 10.14 137.49 7.20
CU 180.53 2.02 0.22 0.00 0.00 0.00 0.00 0.00
DI 0.01 0.60 590.29 552.71 194.64 2.59 204.73 20.15
GWI 4.07 5.31 70.64 0.27 0.00 0.00 0.00 0.00
PE 17.38 527.50 158.44 54.21 2.31 1.94 0.25 0.12
PVC-M 0.00 0.00 189.66 91.61 28.66 0.00 15.26 0.00
PVC-U 2.15 55.45 1719.86 829.82 197.47 12.80 71.68 3.55
RC 0.00 0.00 0.00 0.23 0.06 0.01 7.23 26.95
S 0.01 0.15 8.74 20.91 87.33 0.27 120.14 53.12
CHAPTER 3. PIPE NETWORK HISTORICAL FAILURE ANALYSIS
39
40
40
30
20
10
0
0-50 100-150 200-250 300-350
50-100 150-200 250-300 350-400
Pipe diameter
Figure 3.5: Average Failure rate by diameter over observation period
rate observed for this diameter group.
3.4.5 Soil type

Section 2.6.3 reported that the failure of pipes has been associated with soil
movement, both swelling and shrinkage, specifically that failure rate increases
as the shrink/swell potential of the soil increases. For this reason, soils have
been classified here with regards to their shrink/swell potential (see Appendix
A for more detail on how soils were classified). Soil shrink/swell potential was
categorised as described in Appendix A, the soil codes used in this chapter are
summarised in Table 3.10.
Examination of the average failure rates for in-service assets in each soil
type over the observation period showed that, as expected, failure rate in-
creased as the shrink/swell potential of the soil increased (Figure 3.8(a)). How-
ever, the change in failure rate between ME and EX soils was very small, in
fact the difference between soils was only 0.02 failures/100 km/year, indicating
that it is unlikely that the these two soils are appreciably different with regard
to their affect on pipe failure. Investigation into failure rates of ME and EX
Figure 3.6: Length of assets in each diameter/material group
soils on a yearly basis also found that there was essentially no difference is the
failure rates. For this reason ME and EX soils were combined and treated as
a single soil type, ME+EX (Figure 3.8(b)).
Table 3.10: Soil codes

Soil shrink/swell
Soil code
potential
Very Expansive VE
Mostly Expansive ME
Expansive EX
Slightly Expansive SE
STable ST
To understand how the presence of the soil types actually affects the per-
formance of the two networks the proportions of the supply regions dominated
by each soil were calculated. Figure 3.9 shows the soil distribution after ME
and EX soil have been combined. It can be seen in Table 3.11 that ME+EX
soils dominates around 60% of the area within the supply boundaries of the
two water authorities, with similar proportions existing within both the CWW
and SEWL supply regions. The occurrence of VE soils is disproportionately
high in the CWW supply region. VE soil accounts for around 14% of the
combined supply regions, but accounts for 32% of the CWW supply area and
only 11% of the SEWL supply area. Similarly SE are disproportionately rep-
42
Figure 3.7: Failure rates for assets in each diameter/material group
resented within the SEWL supply region. SE soil accounts for around 16% of
the combined supply boundaries, but accounts for only 4% of the CWW sup-
ply region and 18% of the SEWL supply region. It should be noted that the
area of the CWW and SEWL supply regions are not equal and so percentages
cannot simply be added.
Table 3.11: Percentage of soil types within authority boundaries

Soil code CWW SEWL Combined
VE 32.3% 11.4% 14.4%
ME+EX 62.3% 60.6% 60.9%
SE 4.2% 18.3% 16.3%
ST 1.2% 9.7% 8.5%
70
60
50
40
30
20
10
0
VE ME EX SE ST
Soil type
(a) Initial soil classification
70
60
50
40
30
20
10
0
VE ME+EX SE ST
Soil type
(b) Combined ME and EX soils
Figure 3.8: Average failure rate by soil type over observation period
44
Figure 3.9: Updated soil classifications (ME and EX soil types combined)
These values do not truly represent the situation in these networks as the
entire supply regions of each authority do not contain reticulation mains. More
meaningful statistics were calculated by focusing on the area within the supply
boundary in which pipes were present. The area in which pipes are present
was calculated using a GIS lattice buffer algorithm, based on a 500 metre
lattice buffer across the combined supply regions. The proportions of the
piped area dominated by each soil type were then re-calculated using this
information. The results can be seen in Table 3.12. It can be seen that
CWW has increased and decreased representation of VE and ME+EX soils
respectively when viewed in this manner. The percentage of the area with SE
and ST soils in unaffected. SEWL has reduced representation of both VE and
ME+EX soils, an increase in the representation of both SE and ST soils can
be seen.
3.4.6 Pipe material - soil type interaction

The influence of soil shrink/swell potential was further investigated to establish
how the different pipe materials were affected. The distribution of assets in
Table 3.12: Percentage of soil types where pipes are present

Soil code CWW SEWL Combined
VE 35.7% 9.1% 16.2%
ME+EX 58.8% 49.0% 51.6%
SE 4.2% 27.0% 20.9%
ST 1.2% 14.9% 11.3%
each material/soil group are given numerically in Table 3.4.6 and shown graph-
ically in Figure 3.10. It can be seen that the group with the greatest length
of assets is PVC-U/ME+EX, followed by CI/ME+EX, CI/SE and CI/VE. It
is interesting to note that CI has over 3.5 times more length in-service in VE
soils than the next longest material type.
Figure 3.10: Length of assets in each soil type and material group
Analysis of the failure rates for each material shows that the failure rates for
the oldest materials (AC, CI and GWI) strongly follow the trend of increasing
failure rate with increasing soil shrink/swell potential (Figure 3.11). A similar
trend can be seen to a lesser degree in DI and S. It was expected that RC as a
rigid material would follow a similar trend to that displayed by the other rigid
materials, however this is not seen. The failure rate for RC is similar in both
ST and ME+EX soil types, this may be due to RC pipe being predominately a
material used in larger diameter pipe, over 99% of the length of RC in-service
is in either the 300-350 or 350-400 diameter groups. The high representation of
RC in larger diameter and no clear affect of soil shrink/swell potential supports
the theory proposed by Hu and Hubble (2007). CU, PVC-M, PVC-U and PE
do not seem to follow any discernable trend in relation to the effect of soil
Table 3.13: Length (km) of assets in-service, shown by pipe material and soil type.
The material/diameter groups with the greatest length are highlighted
XXX extra line extra line extra line
X XXXMaterial AC CI CU DI GWI PE PVC-M PVC-U RC S
Soil type XXX
X X
VE 226.75 1150.03 38.74 311.48 29.84 116.78 85.59 259.31 0 66.94
ME+EX 916.42 1303.73 89.38 928.61 14.46 417.74 234.93 1558.07 22.45 109.41
SE 420.48 1258.35 42.16 283.28 28.75 160.45 3.55 620.19 0.12 82.05
ST 423.56 168.43 12.5 42.34 7.23 67.19 1.12 455.2 11.9 32.27
46
shrink/swell potential on failure.
Figure 3.11: Failure rates for assets in each soil and material group
3.4.7 Pipe diameter - soil type interaction

The effect of soil shrink/swell potential was further investigated to establish
how the different pipe diameters were affected. The distribution of pipes
in each diameter/soil type group are given numerically in Table 3.14 and
shown graphically in Figure 3.12. It can be seen that the group with the
greatest length of assets is 100-150/ME+EX, followed by 100-150/SE, 150-
200/ME+EX and 100-150/VE. The 100-150 and 150-200 diameter groups have
lengths far in excess of the other diameters irrespective of soil type.
Analysis of the failure rates for each diameter/soil groups shows that fail-
ure rate increased as soil shrink/swell potential increase for diameter groups
smaller than 250-300 (Figure 3.13). An exception to this trend can be seen for
the 050-100/ME+EX group where the failure rate is seen to drop. The reason
for this drop is due to the high proportion of PE in this diameter and soil,
over 80%, PE represented around 50% of the length in the other soils for this
diameter. The diameters 250-300 and greater showed no clear trend in relation
to increasing soil shrink/swell potential. As noted in Section 3.4.6 this sup-
ports the theory proposed by Hu and Hubble (2007) where larger pipes are less
affected due to higher moments of inertia for larger diameter pipes, reducing
the flexural stress which can develop in the pipe as a result of bending.
48
Table 3.14: Assets in-service, shown by diameter (mm) and soil type by length
(km). The material/diameter groups with the greatest length are highlighted
extra`line extra line extra line
Soil type
```
```
``` VE ME+EX SE ST
Diameter ```
``
0-50 45.92 99.31 50.94 15.74
050-100 104.67 410.73 162.22 76.10
100-150 1248.73 2783.71 1649.34 690.16
150-200 614.53 1378.32 654.15 266.00
200-250 139.09 484.67 196.95 82.53
250-300 7.64 20.43 8.67 5.42
300-350 117.04 344.91 138.93 50.04
350-400 7.83 73.13 38.19 35.77
Figure 3.12: Length of assets in each Soil Type and Diameter Group
3.4.8 Failure type

Different failure types result from different pipe loading conditions as described
in Section 2.5. Failure types were grouped as longitudinal failures (LF), cir-
cumferential failures (CF) and other failures (OTHER), see Appendix A. Fig-
ure 3.14 shows the failure rates for each failure types over the observation
period. It can be seen that OTHER failures occur at twice the rate of CF and
four times that of LF. The reason for the high number of failures allocated as
OTHER is due to the high number of failures for which either no description or
with uninformative descriptions such as ”UNSPECIFIED” were given. These
failures accounted for approximately around 33% of OTHER failures.
Figure 3.13: Failure rates for assets in each Soil Type and Diameter Group
3.4.9 Pipe age
The influence of pipe age on the failure of pipes was discussed in Section 2.6.4.
Whilst the applicability of age alone as a predictor of pipe failure is disputed,
all are expected to deteriorate over time until eventually failure occurs. As no
other indicator of pipe capacity deterioration is available in the data, pipe age
has been used here. The age of assets in-service over the observation period
ranges from those installed in 2006 (0 years old) to those installed in 1856 (150
years old). Figure 3.2 shows the length of assets installed in each year.
To determine whether the deterioration of the assets could be seen in the

failure history of the network, the failure rate was determined based on the
age at the time of failure. It can be seen in Figure 3.15 that the network does
seem to follow the expected trend of increasing failure rate between zero and
approximately 50 years of age. After 50 years of age, the failure rate drops
until approximately 80 years of age. This behaviour is likely to be the result
of a survivor bias, where pipes with lower failure rates are retained within the
network, whilst pipes with higher failure rates are replaced. As no data on
replaced assets were available, the presence of a survivor bias could not be
confirmed.
50
25
20

15
10
0
CF LF OTHER
Figure 3.14: Failure rate of each failure type over the observation period
3.4.10 Pipe age - pipe material interaction

To identify the contribution of the different material types to the observed
trend in failure rate with increasing age, the failure rates of the materials with
the greatest lengths in-service, AC, CI, DI and PVC-U was determined (Figure
3.16). The expected trend of increasing failure rate with age can be seen for
AC, DI and PVC. The expected trend was not seen for CI.
A comparison of the failure rates seen in Figure 3.15 and 3.16 indicate
that CI failures strongly influence the failure rate trend seen in Figure 3.15,
displaying a similar failure rate trend. It should be noted that prior to 1928,
CI made up the majority of the network and so this influence is expected. In
fact, CI accounted for over 98% of the in-service length older than 68 years of
age in 1996 (78 years of age in 2006) and over 99% of the recorded failures for
these assets.
The drop in failure rate exhibited by CI after approximately 50 years of age
and then the steady increase from approximately 60 years of age may indicate
multiple cohorts of pipes in-service. Additionally the observed trend may be
the result of a survivor bias as noted above.
Figure 3.15: Failure rates of assets as they age
Figure 3.16: Failure rates of assets as they age by material

52
3.5 Investigation of seasonal factors influenc-

ing failure rate
This section details the statistical analysis undertaken to determine how sea-
sonal variation in climate can affect failure rates. In order to investigate sea-
sonal variations in failure rate the monthly failure rate is used. The monthly
failure rates within a calendar year are referred to as intra-year failure rates
for the remainder of this chapter. In addition to the network data detailed
in Section 3.2, daily climate data were obtained from the Australian Bureau
of Meteorology (BOM). These data were acquired for ten weather stations
distributed within the pipe networks.
3.5.1 Intra-year variation of failure rates

The intra-year variation of failure rates in each year of the observation period
is shown in Figure 3.17. A peak in failure rates is clearly observed at the start
of the year, its seasonal occurrence (corresponding to mid to late summer) is
similar to the peak observed by Hudak et al. (1998) for southern USA.
The summer peak seen in Figure 3.17 can be seen most clearly in 1997, 2001
and 2003 with failure rates peaking above 8.5 failures/100 km/month (equiv-
alent to over 102 failures/ 100 km/year) in all three years. The highest fail-
ure rate was recorded in February 1997, exceeding 9.5 failures/100 km/month.
However, the summer peak is not observed every year and significant difference
in the variation of intra-year failure rates is present over the observation period.
For example 1999 and 2002 experienced relatively consistent intra-yearly fail-
ure rates, varying by around 2 failures/100 km/month. This contrasts strongly
with 1997, 2001 and 2003, where intra-yearly failure rates varied by up to 7.5
failures/100 km/month
3.5.2 Climatic parameter correlation investigation

Several authors including Habibian (1994), Hu and Hubble (2007) Hudak et al.
(1998) and Rajani and Zhan (1996) identified the importance of climatic pa-
rameters in understanding intra-yearly variation in failure rate. For this reason
a range of climate parameters were investigated to identify which correlated
best with failure rate. The literature identifies temperature and soil moisture
content as being significant climate related factors affecting pipe failure. Fol-
lowing this approach, thirteen monthly summary parameters are included in
Figure 3.17: Intra-year failure rate variation
this investigation: minimum antecedent precipitation index (API), net evap-

oration, minimum temperature, maximum API, maximum temperature, av-
erage evaporation, average evapotranspiration, average API, average rainfall,
net evapotranspiration, monthly evaporation, monthly rainfall and monthly
evapotranspiration.
API is calculated as a daily time series using Equation 3.3. As the initial
value chosen for the calculation of API affects subsequent results, API is cal-
culated starting five years prior to the observation period to negate this affect.
The initial value, AP In=0 of 0.0 is used in these calculations.
AP In = k × AP In−1 + Pn , n = 0, 1, 2 . . . m (3.3)
where AP In is the antecedent precipitation index at day n, Pn is the precip-

itation recorded for day n and k is a coefficient set as 0.85 in line with Zhou
et al. (2001b).
The monthly summary parameters for temperature and API are determined
as the minimum, maximum and average values during each month. Monthly
rainfall, monthly evaporation and monthly evapotranspiration are the sum of
daily rainfall, potential evaporation and potential evapotranspiration in each
month respectively. Average rainfall is calculated as the monthly rainfall di-
vided by the number of days with recorded rainfall. Average evaporation and
average evapotranspiration are calculated from the monthly evaporation and
54
monthly evapotranspiration divided by the number of days in the month. Net

evaporation and net evapotranspiration are calculated from the monthly evap-
oration and monthly evapotranspiration respectively less the monthly rainfall.
Each of the thirteen monthly summary parameters were calculated for each
ten weather stations for every month over the ten year observation period. This
created a total of 1200 month-weather station data points for each monthly
summary parameter. Pipes are assigned the climate data of the closest weather
station to ensure any spatial variation is accounted for.
To determine the level of correlation between monthly summary parameters
and failure rate, the range of values calculated for each monthly summary
parameter was broken into discrete intervals. The exposed length and number
of failures of all month-weather station data points within each interval are
summed and the failure rate of each interval calculated. This allows temporally
and/or geographically disparate data to be combined.
Different trendlines were then fitted to the data and the most appropriate
was determined based on the strength of the correlations. The coefficient of
determination (R2 ) was used as the metric to compare the strength of corre-
lations. ANOVA (analysis of variance) was then conducted on the parameters
with the best results to statistically determine the reliability of the correlation
observed.
Pipe failure rates were calculated with reference to discretised monthly
summary parameters to detect potential correlations. A range of trendlines
were then fitted to the data. An exponential trendline was found to provide
the best fit.
Using an exponential trendline minimum API and net evaporation were
both found to correlate strongly with failure rate with R2 values of 0.78 and
0.74 respectively. Figure 3.18 and Equation 3.4 show the histogram and fit-
ted trendline for minimum API respectively. Equation 3.5 shows the fitted
trendline for net evaporation. An ANOVA of the data for minimum API and
net evaporation indicated that there is indeed a dependence of failure rate on
both parameters. The results indicated that there is negligible probability (<
0.0001) that the variation observed could be explained without the use of the
trendlines determined during the regression analysis. As minimum API was
the monthly summary parameter which best correlated with failure rate it will
be used to represent climate for the remainder of this chapter.
The correlation between minimum API and failure rate indicates that as
minimum API decreases failure rate increases. API can be used as a measure
of soil moisture content (Blanchard et al., 1981), where the moisture content
of the soil increases as API increases. From this relationship it can be inferred
that as the moisture content of the soil decreases failure rate increases. This
result is in agreement with the qualitative observations made by Hudak et al.
(1998), Kleiner and Rajani (2000) and Newport (1981), for southern USA,
southern Australia and the UK respectively. The relationship between soil
and minimum API is investigated and discussed below.
Following this result the cause of the variation in the amplitude of the
observed peak in failure rate during mid to late summer, Figure 3.17, was
investigated. It was found that in the months January through March for
years where peaks in failure rates were observed, low values of minimum API
were recorded (around 50% lower than the average). Conversely higher values
of minimum API (around 50% higher than the average) were calculated in
those years in which no peak was observed.
FR = 3.34 ✕ e - 0.089 ✕ MinAPI

R 2 = 0.78
Figure 3.18: Histogram of pipe failure rate depending on Minimum API
F ailureRate = 3.34 × e−0.089×M inAP I R2 = 0.78 (3.4)
F ailureRate = 2.72 × e0.005×N etEvaporation R2 = 0.74 (3.5)

56
3.5.3 Interaction between climate and other factors

As the failure of buried pipes is strongly influenced by both static and dynamic
factors, the relationship between these factors (pipe material, pipe diameter,
soil shrink/swell potential and observed failure type) and climate were investi-
gated. Climate is represented by minimum API as it is the monthly summary
parameter which best explained the intra-year variation in failure rates. The
effect of climate on the failure rates for each factor was investigated by plot-
ting histograms for each category within each factor. A category refers to the
different values within each factor, e.g. CI and PVC-U are categories of pipe
material. A trendline is then fitted to the histogram for each category.
The presence of an effect is determined using two tests based on the trend-
lines for each category within each attribute. An effect is said to be present if
either one or both of the following tests indicated an effect due to climate.
Test 1: The relative values of maximum failure rate for each category
determined from the trendlines are compared to the relative failure rates of
that category without reference to the effect of climate. Where the maximum
failure rate determined from the trendline is higher or lower than expected,
an effect was said to be present. For Test 1 where an effect is present this
indicates that when subject to the most deleterious climatic conditions, pipes
within that category exhibit more/less failures due to the presence of these
conditions than expected.
Test 2: The trendline gradient of each category is compared to that of
the trendline fitted to the entire dataset. Where the gradient is higher or
lower than that of the entire dataset an effect is said to be present. For
Test 2 where an effect is present this indicates that pipes within the category
under investigation exhibit higher/lower than expected sensitivity to changes
in climate.
For the purpose of this analysis it is assumed that no pipes are subject to
special conditions. All pipes are assumed to be subject to the same loading
and environmental conditions.
3.5.3.1 Effect of climate and pipe material on failure
Due to the number of pipe materials present in the data, only the four materials
(CI, AC, PVC-U and DI) with the greatest in-service length were included in
the analysis. These materials accounted for over 80% of the pipe length, see
Table 3.15. The presence of an effect was determined using two tests detailed
in the methodology section. An effect was said to be present if either one or
both of these tests indicated an effect.

The parameter affect investigation for pipe material and minimum API
indicated no affect for Test 1 for all materials. Test 2 did indicate an effect for
CI and PVC-U, a negative result was found for AC and DI. The failure rate
of CI was found to increase faster than expected as minimum API decreased;
whilst the failure rate of PVC-U increased slower than expected as minimum
API decreased. For all materials, the failure rate was found to increase as
minimum API decreases.
As discussed in Chapter 2, the failure of buried pipes occurs when applied
stress exceeds pipe capacity. Therefore two possible situations exist. The
increase in pipe failure rate occurs due to an increase in applied stress and/or
an increase in the rate of reduction of pipe capacity.
As all pipes are assumed to be subject to similar conditions, the higher
and lower sensitivity of CI and PVC-U respectively is a result of the materials
response to these conditions. To determine the most likely situation, the effect
of changing minimum API was investigated in relation to pipe capacity. As AC
and DI did not show an interactive affect with minimum API the behaviour
of these materials has not been investigated.
The structural capacity of CI pipes reduces due to corrosion. The rate of
corrosion is dependent on soil corrosivity. The corrosivity of a soil increases
as soil resistivity decreases (Sadiq et al., 2004) and soil resistivity decreases
as soil moisture content increases (Zhou et al., 2001a). As these conditions
are unlikely to occur in response to a reduction in minimum API (indicating
a decrease in soil moisture content), it is therefore likely that the increased
failure rate results from an increase in pipe stress rather than a reduction in
capacity.
The capacity of PVC-U pipes reduces as a result of slow crack growth from
defects in the pipe wall, the rate of which is dependent on the defect size and
pipe stress conditions (Burn et al., 2005). As initial defect is determined at the
time of or prior to pipe installation and is not related to minimum API, any
variation in the rate of pipe capacity reduction is dependant on pipe stress.
Consequently, the increased failure rate is most likely due to an increase in
pipe stress.
The increase in failure rate for these materials is almost certainly due to
an increase in pipe stress. Following this conclusion, the higher sensitivity of
CI results from the development of higher than expected pipe stresses with
decreasing minimum API, with the opposite true for PVC-U.
58
Figure 3.19: Intra-year failure rate variation by Material
3.5.3.2 Effect of climate and pipe diameter on failure
Due to the number of pipe diameters present in the data only the four diameter
groups (50−<100mm, 100−<150mm, 150−<200mm and 200−<250 mm) with
the greatest in-service length were included in the analysis. These diameter
groups accounted for over 80% of the pipe length, see Table 3.15. The presence
of an effect was determined using two tests detailed in the methodology section.
An effect was said to be present if either one or both of these tests indicated
an effect.
The results from both Test 1 and Test 2 did not indicate an effect of climate
with diameter.
It should be noted that this does not mean that there is no variation in
failure rates between diameter groups. Only that the variation seen in response
to minimum API is consistent with the expectations tested in Test 1 and Test
2.
Table 3.15: Exposed length and number of failures of material and diameters
included in analysis
% of total Number of % of total
Category Length (km)
length failures failures
CI 3881 32% 25898 65%
Asbestos Cement 1987 17% 8477 21%
PVC-U 2893 24% 1673 4%
Ductile Iron 1566 13% 1913 5%
50-¡100 754 6% 2358 6%
100-¡150 6372 53% 27213 69%
150-¡200 2913 24% 7644 19%
Figure 3.20: Intra-year failure rate variation by Diameter
3.5.3.3 Effect of climate and different soil types on failure
The effect of minimum API on the failure rate of pipes in different soil types
was investigated for all four soil types identified during the analysis; very
expansive, expansive, slightly expansive and stable. The presence of an effect
was determined using two tests detailed in the methodology section. An affect
was said to be present if either one or both of these tests indicated an effect.
Test 1 and 2 both indicated an effect for very expansive soils; very expansive
soils were found to experience a higher than expected maximum failure rate
and the failure rate for very expansive soils increased faster than expected
with a decrease in minimum API. Test 1 and 2 both indicated no effect for
the remaining soil types. These results indicate an effect of minimum API on
the failure rate of pipes in different soil types, with minimum API strongly
affecting the failure rate of pipes buried in very expansive soils.
As discussed in Chapter 2, the failure of buried pipes occurs when applied
stress exceeds pipe capacity. It was discussed above that the increase in failure
rates are likely to be due to an increase in the stress as minimum API decreases.
The higher sensitivity and the higher than expected maximum failure rate of
pipes buried in very expansive soils indicates a situation where these pipes
are subject to higher increases in stress as minimum API decreases than pipes
buried in other soil types.
Following this the stress in buried pipes increases as soil moisture content
decreases. As a reduction in moisture content of expansive soils results in the
shrinkage, this shrinkage is the cause of the increased stress. Furthermore,
as very expansive soils are by definition strongly affected by moisture change
60
these soils would impose harsher conditions on pipes than less expansive soils.
Figure 3.21: Intra-year failure rate variation by Soil
3.5.3.4 Effect of climate on observed failure type
The effect of minimum API on failure type was investigated for circumferential
and longitudinal failures. Other failure types were not included. The presence
of an effect was determined using two tests detailed in the methodology section.
An effect was said to be present if either one or both of these tests indicated
an effect.
Results from Test 1 did not indicate an effect. Test 2 indicated an effect
of climate on circumferential failures and no effect of climate on longitudinal
failures.
Analysis of intra-year failure rates for each failure type revealed that failure
rates for longitudinal failures were relatively consistent over the course of the
calendar year, whilst circumferential failures showed a clear peak. The monthly
failure rates for longitudinal failures varied by around 30% from the average.
In contrast circumferential failure rates showed a clear peak toward the end of
summer; varying by around +100% from the average in summer and by around
-60% in winter. Capacity reduction should result in an increase in failures
irrespective of the type. Whereas, the summer increase in circumferential
failures indicates an increase of the associated loading types at this time.
Circumferential failures result from longitudinal stresses caused by flexu-
ral loads and/or restrained thermal contraction (Rajani et al., 1996). As the
increase in failures occurs in summer the latter cause can be excluded. There-
fore the increase in circumferential failures is most likely due to an increase in
flexural stress.
Figure 3.22: Intra-year failure rate variation by Failure Type
3.6 Conclusions
The analyses presented in this chapter determined the trends in pipe failure
with respect to pipe material, pipe diameter, surrounding soil type, pipe failure
type and pipe age. Pipe material, pipe diameter and the shrink/swell potential
of the surrounding soil were all found to strongly influence the occurrence
of pipe failure. The relative ranking of pipe material with regard to failure
rate was found to be in line with the data reported in the literature for CI,
AC, PVC and DI. The failure rate trends observed for pipe diameter and
soil shrink/swell potential were found to be consistent with the behaviour
reported in the literature. Pipe failure rate was found to increase as pipe
diameter decreased to a maximum for pipe with a nominal diameter of 100
mm. The failure rate of pipes with smaller diameters was found to decrease,
this change in trend is believed to be the result of the change in dominant pipe
material from CI to PE for these smaller diameter pipes. The failure rate of
pipes was found to increase as the shrink/swell potential of the surrounding
soil increased. The effect of pipe age on observed failure rate was investigated
for CI, AC, DI and PVC-U. A trend of increasing failure rate with age was
found for AC, DI and PVC-U. CI was found to show an increasing failure rate
to around 50 years of age. Failure rate then decreased before again increasing
after approximately 60 years of age. This behaviour is believed to be the result
of the CI pipe material consisting of two distinct types of CI.
62
Investigation into the seasonal variation in pipe failure number found that
pipe failures peak in Melbourne between mid summer to late autumn, with the
highest failure rate occurring in February. The failure rates in these months
were observed to vary significantly between years, causing the summer peak
in failure rates to be strongly evident in some years and scarcely observable
in others. Monthly pipe failure rate was found to correlate strongly with the
minimum antecedent precipitation index (minimum API). Failure rate was
found to increase as minimum API decreased. Years in which low values of
minimum API were recorded during mid summer to early autumn (around
50% lower than the average) were found to have an strongly evident peak in
the failure rate. Years in which high values of minimum API were recorded
during mid summer to early autumn (around 50% higher than the average)
were found to not have an observable peak in the failure rate.
The effect of minimum API on failure rate was found to vary depending on
pipe material and the surrounding soil type. Change in minimum API were also
found to influence the observed failure type. Investigation into the relationship
between minimum API and pipe material indicated that, the observed increase
in pipe failure rate is likely to be related to an increase in pipe stress. The
relationship between the observed failure type and minimum API indicated
that, this pipe stress is likely to be a flexural stress. The relationship between
the surrounding soil type and minimum API indicated that the magnitude of
this flexural stress increases as soil shrink/swell potential increases.
Based on these findings, the following hypothesis was formulated: that
the increase in the number of failures in buried water reticulation pipes in
Melbourne, Australia during mid summer to early autumn occurs as a result
of soil shrinkage.
Chapter 4
Field instrumentation of an
in-service pipe in a high
shrink/swell potential soil
4.1 Introduction
This chapter reports Stage two of the research. The analysis presented in
Chapter 3 allowed the hypothesis that, soil shrinkage is the cause of the sea-
sonal variation in pipe failure rate to be formulated. This chapter presents a
detailed field study conducted to establish the causes of seasonal variation in
pipe failure numbers to investigate this hypothesis.
The field study was undertaken on an in-service buried water reticulation
pipe. The pipe and its surrounding environment were monitored to determine
the response of a pipe in the field over a two year period by recording pipe
strain, soil water content, soil temperature, soil suction, soil movement and
weather conditions.
The aim of this work was to use field data to determine the causes of
seasonal variation in pipe failure numbers and thus to confirm or refute the
hypothesis that the increase in pipe failure numbers during mid summer to
early autumn occurs as the result of soil shrinkage.
Section 4.2 details the site selection and the results of a geotechnical in-
vestigation of the site prior to the installation of instrumentation. Section
4.3 describes the sensors and instruments installed and the calibration of that
equipment. Section 4.4 details the field work undertaken to instrument the
reticulation pipe and its surrounding environment, and Section 4.5 reports the
analysis of the collected data.
63
64
The field study detailed in this chapter was a collaborative study in concert
with a second PhD student. The analysis of the raw data to calculate pipe
stress from the strain gauge data does not form part of this thesis. Details
of that analysis can be found in Chan (InPrep), a forthcoming PhD thesis.
The research presented in this chapter has been presented in two papers at an
international peer-reviewed conference (see List of publications and Appendix
??).
4.2 Site selection and investigation

In order to select the site which will be the focus of this detailed field study, a
comprehensive evaluation of possible sites was undertaken. Several sites were
initially selected following the set of criteria detailed in Section 4.2.1. The
geotechnical properties of these sites were then investigated to identify the
most appropriate site. The results of the geotechnical investigation for the
selected site are given in Sections 4.2.2 and 4.2.4.
4.2.1 Site selection criteria

The criteria considered during selection of the field study site are listed below.
These criteria ensured the selected study site was effective, safe and convenient
for excavation, instrumentation and data-logging.
• Located in the CWW supply region;
• Located in an area of high failure rates history;
• Contain a buried cast iron (CI) water reticulation pipe with nominal
diameter between 100 mm and 150 mm (referred to as the pipe for the
remainder of this chapter). These pipe attributes were selected as they
identified the cohort of pipes most affected by seasonal climatic changes;
• Located within a region with high shrink/swell potential soil, with a

minimum soil depth of 1.5 m;
• The pipe must be run under a nature strip (pervious surface) and drive-
way (or similar impervious surface);
• No previous pipe failures to have occurred within the instrumentation

length;
CHAPTER 4. FIELD INSTRUMENTATION 65
• Wide nature strip to allow for instrumentation;
• Relatively flat ground surface to avoid the effects of sloping ground and
risk of potential flooding;
• Sufficiently clear of other utilities such as gas, power, telecommunica-

tions, storm water and sewer;
• No trees in the nature strip close to the instrumentation locations; and
• A quiet area with relatively low traffic flow.
A large number of sites were screened using driving and walking surveys
using the above criteria. Ten sites were selected during these initial surveys.
From these ten sites the two most suitable sites were chosen for additional
investigation to determine soil depth and site specific properties. Soil investi-
gations were done by hand auguring and undisturbed push tube sampling by
a drill rig to determine the depth of soil at these two sites. The site chosen for
instrumentation is located in Altona North, VIC (Figure 4.1). This site has a
clay layer depth of more than the minimum 1.5 m below the ground surface.
The other site failed this selection requirement.
4.2.2 Site soil bore hole analysis

Undisturbed push tube sampling was conducted to determine the depth of soil
at the site, to identify any variation in soil type with increasing depth and to
provide samples for further testing. Figure 4.2 shows the observed vertical soil
profile at the field study site. It can be seen that the first 0.4 m is surface soil
containing organic material, including plant (grass) roots. Below 0.4 m to 2.0
m the soil is inorganic residual clay covering basalt rock.
4.2.3 Site geometry

The total length of the nature strip at the field study site was 23.8 m and the
width is 4.8 m, giving sufficient access to provide instrumentation in the soil
and the pipe. In addition to the pipe to be instrumented, which was located
3.16 m from the property boundary, a 100 mm gas pipe was situated 2.4 m from
the property boundary and a storm water pipe 0.6 m from the road edge. Power
and telecommunications lines were located 6 m overhead. Surface objects
existing on the nature strip were the power pole and a small tree, neither of
which interfered with the instrumentation or subsequent measurements.
66
144°30' 145°00'
MAP SCALE 1:250 000
CWW Supply Region
metres 5000 2500 0 5 10 15 20 25 kilometres
Figure 4.1: Geological map of the CWW supply area showing the location of
the Altona North field instrumentation site Rixon (1973)
4.2.4 Soil classification tests

Soil collected during push tube sampling was tested to confirm the shrink/swell
potential of the soil. After collection, samples were sealed in plastic bags
and stored in a constant temperature room until needed for testing. All soil
tests were performed in accordance with the relevant Australian Standards.
The preliminary preparation of the soil samples was performed in accordance
with the requirements of the AS1289.1.1 2001 (Standards Australia, 2001).
The results of the soil classification tests are presented in the remainder of
this section. The soil samples used in the following classifications tests were
obtained from depths between 0.5 m and 0.8 m unless otherwise noted.
4.2.4.1 Particle size analysis
The particle size analysis for particle sizes down to 75 µm was conducted by
dry sieving following AS1289.3.6.1 (Standards Australia, 2009b). Soil par-
DRILLING METHOD
DEPTH (m)
INFERRED STRATIGRAPHY
0.00
Surface Soil: MH silty clay, dark ash, with

fibers, dry to moist
0.40
Push Tube
CH inorganic clay of high plasticity, dark ash

to light ash, moist, stiff to very stiff
(Residual basaltic clay)
2.00
Rock Face
Basaltic Rock
Figure 4.2: Bore hole log data from the Altona North field instrumentation
site (Note: the ground water table was not observed during sampling)
ticles smaller than 75 µm were prepared by the wet sieving method in ac-
cordance with AS1289.3.6.2 (Standards Australia, 1995b). The particle size
analysis for particle smaller than 75 µm was then conducted in accordance
with AS1289.3.6.3 (Standards Australia, 2003). The combined results of the
particle size analysis are shown graphically in Figure 4.3.
4.2.4.2 Plastic and liquid limits
The plastic and liquid limits were determined in accordance with AS1289.3.2.1
(Standards Australia, 1995a) and AS1289.3.1.1 (Standards Australia, 2009a)
respectively. The plastic limit is the water content where soil starts to exhibit
plastic behaviour. The liquid limit is the water content where a soil changes
from plastic to liquid behaviour. In both tests, the water content was deter-
mined by the oven drying method (AS1289.2.1.1 (Standards Australia, 2005)).
The plastic and liquid limits are used to classify the soil (see Section 4.2.4.3).
The plastic and liquid limits determined for the field study site are 21.8% and
70.2% respectively.
68
Particle Size (mm)
Figure 4.3: Particle size distribution of soil sample from Altona North field
study site
4.2.4.3 Plasticity index
The plasticity index of a soil is the difference between its plastic and liquid
limits. The plasticity index of the soil at the field study site was found to be
48.4%. The plasticity index and liquid limit can be used for soil classification
using the plasticity chart in accordance to AS1726 (Standards Australia, 1993).
The soil at the field instrumentation site is classified within the region of
inorganic clays of high plasticity.
According to the study of soil shrink/swell potential on the basis of At-
terberg limits (Holtz and Gibbs, 1956, Seed et al., 1962), soils with plasticity
index greater than 35% can be considered as having a high shrink/swell po-
tential. The high plasticity index of 48.4% obtained from the field study site
indicates that the soil is considered to have a high shrink/swell potential.
4.2.4.4 Linear shrinkage
Linear shrinkage was determined from the linear shrinkage test AS1289.3.4.1
(Standards Australia, 2008). The measured linear shrinkage value for the
Altona North field instrumentation site was 16%.
4.2.4.5 Specific gravity, dry density and void ratio
The specific gravity (Gs ) of soil at the field study site was determined using
the AccuPyc 1330 Pycnometer for a disturbed sample. The sample was broken
by hand and oven dried for 24 hours at 105 °C in order to remove water.
The sample was then crushed by hand using a mortar and pestle, and passed
through a 1.18 mm sieve, then returned to the oven for a further 24 hours to
ensure all water had been removed. Once dried, the sample was removed from
the oven allowed to return to room temperature in a desiccator containing
silica gel to ensure no water was absorbed prior to testing.
A representative sample was obtained by quartering until the required size
was obtained. The mass of the sample was measured to an accuracy of 0.0001
g. The pycnometer then measured the volume of the specimen (ten repeat
measurements were conducted and the average taken) and the density calcu-
lated. The specific gravity was calculated as 2.6206 at 24.1 °C.
The dry density of the soil was determined using undisturbed specimens col-
lected during push tube sampling. Specimens of known volume were obtained
at 100 mm increments and the mass determined. The water content of adjacent
soil samples was then determined by the oven drying method (AS1289.2.1.1
(Standards Australia, 2005)). The dry density of each specimen was then cal-
culated. The void ratio was calculated at 100 mm increments from the dry
density and Gs , Figure 4.4 shows the results of this testing.
4.2.4.6 Swelling properties
The swelling properties of soil samples collected from the Altona North field
instrumentation site were determined by the oedometer test in accordance
with D4546 (ASTM International, 2003). Three tests were undertaken with
soil samples of different dry density.
Several types of apparatus are available for measuring shrink/swell prop-
erties. Figure 4.5 shows two types of apparatus, Figure 4.5(a) is a typical
oedometer in which the specimen is loaded by the weight from a hanging lever
arm. Figure 4.5(b) shows a new apparatus, manufactured in the Monash Uni-
versity workshop. This apparatus is smaller in size and has a digital dial gauge,
which can be connected with a computer to continuously record specimen dis-
placement. It can be used to measure free swelling potential of soils, but is
limited by the maximum pressure that can be applied to the sample during
testing.
Undisturbed soil samples collected from the field study site were cut to cre-
70
Dry density (g/cm3)

0 0.3 0.6 0.9 1.2 1.5 1.8
200
400
600
Depth (mm)
800
1000
1200
1400
1600
1800
2000
0 0.2 0.4 0.6 0.8 1.0 1.2
Dry density
Figure 4.4: Dry density and void ratio of undisturbed soil collected from field
site
ate specimens suitable for testing. The top and bottom of the specimens were
trimmed to create parallel surfaces for axial loading in the oedometer. The
specimen was then placed in the oedometer, inundated with water and allowed
to swell freely (subject to a nominal pressure of 1 kPa). After free swelling
was completed the specimen was compressed by applying a 5 kPa load and the
resulting vertical displacement measured over time and allowed to stabilise.
The applied pressure was then doubled (to 10 kPa) and the vertical displace-
ment again allowed to stabilise. This process of doubling the applied pressure
and stabilisation was continued until the specimen was recompressed past its
original void ratio. The vertical displacements were continuously recorded by
a dial gauge. Figure 4.6 shows results of one swelling test.
In this form of testing, the soil swelling pressure is defined as the pres-
sure required to compress the specimen back to its original void ratio. It is
determined as the pressure corresponding to zero swelling strain on the recom-
pression curve. The swelling pressure for soil at the Altona North field study
site is approximately 660 kPa.
This means that, based on these tests, if the soil under the pipe swells,
Lever Arm
Loading Plate
Dial Gauge
Dial Gauge
Oedometer Cell
Oedometer Cell
(a) Typical oedometer (b) Monash designed oedometer
Figure 4.5: Oedometers used for determining shrink/swell properties
is the pipe does not move (i.e. fully restrained conditions giving no vertical
strain) and assuming full lateral restraint, the soil is able to exert a maximum
swelling pressure of 660 kPa on the pipe bottom. If the pipe does move the
actual stress exerted on the pipe after relaxation can be computed from Figure
4.6 referring to the corresponding soil strain. However, it should be noted that
soil swelling characteristics vary with the initial soil water content and dry
density for the same soil type. The initial soil water content and dry density
of the specimen whose results are shown in Figure 4.6 were 23% and 1.39
g/cm3 respectively in this case.
4.2.4.7 Mineralogy analysis
A soil sample collected from the field instrumentation site was sent to the
Mineralogical and Geochemical services at the CSIRO in South Australia for
quantitative mineralogy analysis. Prior to dispatching, the sample was oven
dried to a constant mass, then crushed to powder form. At the CSIRO, the
soil particles were reduced to sub-micrometre sizes by wet grinding using a
McCrone micronizing mill with ethanol. The analysis was undertaken using the
commercial package SIROQUANT for X-ray diffraction analysis, the results
72
35
Initial dry density = 1.39 g/cm3
30 Initial water content = 23 %
25
Swelling Strain (%)
20
15
10
Swelling pressure=660 kPa
5
-5
0 100 200 300 400 500 600 700 800
Applied Stress (kPa)
Figure 4.6: Soil swelling strain vs. applied pressure for a sample from field
instrumentation site
are shown in Table 4.1. The significant presence of clay minerals, including
smectite and kaolin imparts high shrink/swell potential to the soil.
Table 4.1: Mineralogical content of soil sample from field site

Quartz 58%
Albite 2%
Orthoclase 3%
Kaolin 2%
Smectite 31%
Calcite 3%
Halite <1%
Ilmenite -
Anatase <1%
4.2.4.8 Summary
A summary of the tests described above is given in Table 4.2. The classification
test results have provided information on the soil properties that confirm the
applicability of the site.
Table 4.2: Summary of the soil classification test results
Colour Light brown / beige

Linear shrinkage 16.0%
Liquid limit 70.2%
Plastic limit 21.8%
Plasticity index 48.4%
Soil group Inorganic clays of high plasticity
4.3 Sensors and instrumentation equipment

To monitor the behaviour of the pipe at the field site, a 63 sensors were attached
directly to the pipe and installed into the soil surrounding the pipe. The sensors
used and their purposes are as follows:
• 12 biaxial strain gauges attached to the pipe surface for measurement of

pipe deformation
• 15 thermocouples for measurement of soil temperature
• 15 thermal conductivity (Campbell type) sensors for measurement of soil

matric suction
• 15 soil volumetric water content sensors (Theta type)
• 1 rod extensometer with four anchors to monitor the vertical soil move-
ment at different depths
• 2 earth pressure cells to measure the soil pressure exerted on the pipe
• 1 pressure gauge to measure pipe water pressure
• 1 temperature gauge to measure pipe water temperature
• 1 weather station for the measurement of the air temperature, rainfall,

humidity, wind speed and solar radiation.
In addition three sacrificial anodes were connected to the instrumented pipe

sections to reduce the pipe corrosion in order to prolong the lifespan of the
gauges.
The following section details the sensors installed at the field site and any
calibration undertaken according to the manufacturer’s specifications.
74
4.3.1 Strain gauges

Strain gauges were installed to measure the deformation response of the pipe.
General purpose 3-wire waterproof biaxial strain gauges (KFW-5-120-D16-11
from Kyowa, Japan) were selected. These gauges are thermally compensated
for materials with a thermal expansion coefficient of 11 µε/°C. Thermally
compensated strain gauges measure no strain due to temperature change if
the material has the same expansion coefficient as the compensated thermal
coefficient of the strain gauge.
4.3.2 Earth pressure cells

The pressure applied to the pipe by soil swelling and shrinkage was measured
using vibrating wire earth pressure cells (Geokon model 4800) with 1 MPa
capacity. The earth pressure cells were installed at locations directly beneath
the pipe. The earth pressure cells were calibrated by applying known loads
and recording the output to produce a calibration curve. The earth pressure
cells were calibrated between steel plates in an attempt to provide a even
distribution of the applied loads across the surface of the cell
4.3.3 Thermocouples
Thermocouples (type T thermocouple burial sensors 105T-L from Campbell
Scientific Inc.) were installed in the soil surrounding the pipe to monitor
temperature at various depths. All the thermocouples were tested in the lab-
oratory by immersion in water at a known temperature prior to installation in
the field.
4.3.4 Matric suction sensors

Matric suction sensors (model 229 from Campbell Scientific Inc.) were installed
in the soil surrounding the pipe to measure the suction of the soil at various
depths. The suction sensors are of the heat dissipation type and consist of
a hypodermic needle housing a thermocouple and a heating element, encased
in a porous ceramic tip. When the sensor is activated the heating element
is supplied with an electric charge (50 mA) for 24 s and the increase the
temperature within the sensor measured (Wightwick, 2010). The temperature
change resulting from the electric charge is dependent on the surrounding water
content of the ceramic (ICT, 2007). The sensor is designed in such a way that
the water content of ceramic tip is in equilibrium with the surrounding soil
suction, and with calibration, measures the matric suction of the surrounding
soil.
4.3.4.1 Calibration of matric suction sensors
The calibration of matric suction sensors was undertaken within a pressure

cell using clay soil collected from the field study site prior to instrumentation.
The soil was mixed with water to a water content close to the liquid limit
and compacted into a metal ring placed on top of a ceramic plate, with an
air entry value of 15 bar, inside the pressure cell as shown in Figure 4.7. The
matric suction sensors were inserted into the soil so that the porous ceramic
tip containing the heating device was fully enclosed by soil. The sensor cables
exited the pressure cell through holes in the steel top plate. The calibration
began at atmospheric pressure until the sensor readings stabilised. The air
pressure was then increased to 40 kPa, 50 kPa, 150 kPa and 300 kPa allowing
the sensor readings to stabilise at each stage. Equilibrium times varied from
approximately 1 week, to up to 3 weeks. No cracking of the clay was observed
during calibration. Figure 4.8 shows the calibration curves developed for the
sensors. Exponential trendlines were used to convert the temperature output
of the sensors into suction. Due to the design of the top plate and time
limitations only 10 out of 15 matric suction sensors could be calibrated, the
uncalibrated sensors were assigned the average calibration characteristics.
4.3.5 Soil water sensors

Soil water content sensors (model ML2x manufactured by Delta-T Devices
Ltd.) were installed in the soil surrounding the pipe to measure the water
content of the soil at various depths. The soil water sensors were capable of
measuring volumetric water content of soil based on the change of dielectric
constant caused by differences in soil water. As a result of this, the output
from the sensor could vary between soil types. Therefore the sensors were
calibrated using soil samples collected from the Altona North instrumentation
site.
4.3.5.1 Calibration of soil water sensors
Calibration was undertaken by compacting soil to a known dry density at dif-

ferent water contents into plastic containers. The pins of sensors were inserted
76
Figure 4.7: Calibration of suction sensors with pressure cell
fully into samples and the sensor output recorded. The gravimetric water con-
tent of the samples was measured by oven drying (AS1289.2.1.1 (Standards
Australia, 2005)). This converted to the volumetric water content using the
dry density. The output voltage from the soil water sensor was then correlated
with the volumetric water content.
Figure 4.9 shows the calibration plot of all 15 sensors. The relationship
of measured volumetric water content and sensor’s output (corrected using
the manufacturers calibration constant) can be approximated by two straight
lines, the first for sensor output below 44.8% and the second above 44.8%. The
equations and applicable ranges are shown in Figure 4.9.
4.3.6 Pipe water pressure and temperature gauges
The pipe water pressure and temperature were monitored by pressure and tem-
perature gauges (models SITRANS P 7MF1580 and SITRANS TF2 7NG3140
pressure and temperature gauges respectively manufactured by Siemens Ltd).
The ranges of pressure and temperature gauges were 0 to 10 Bar and -50
to + 200 °C respectively. The manufacturer’s calibration of the gauges was
checked by applying a pressure and immersion in water of known temperature
as appropriate.
(x/0.27939)
Y = 0.03453 e
300
(x/0.28646)
Y = 0.0337 e
(x/0.32068)
Y = 0.0641 e Campbell-1
250
Y = 0.0406 e
(x/0.2861)
Campbell-2
Y = 0.0333 e
(x/0.27531)
Campbell-3
200 Y = 0.0515 e
(x/0.33514)
Campbell-4
Campbell-5
Suction (kPa)
(x/0.29207)
Y = 0.02139 e
Campbell-6
150 (x/0.30357)
Y = 0.02534 e
(x/0.30855) Campbell-8
100 Campbell-9
(x/0.28229)
50
0.8 1.2 1.6 2.0 2.4 2.8

o
Output (Δ C)
Figure 4.8: Calibration of suction sensors
4.3.7 Rod extensometer
A custom-built rod extensometer (model 4000 manufactured by Geotechnical

Systems Australia Pty. Ltd.) was installed in a borehole backfilled with a
weak cement bentonite mix to monitor the sub-soil movement. The rod exten-
someter consisted of four hydraulic inflatable anchors at 0.5 m, 1.0 m, 2.0 m
and 3.0 m from the reference head (Figure 4.10). The reference head, anchors
1, and 2 were expected to move with soil, while anchors 3 and 4 were fixed in
the rock. Transducers in the reference head measure relative soil displacement
between each anchor and the reference head.
The readings of the transducers attached to anchors 3 and 4 give the move-
ment of the reference head with respect to the rock (ideally these two readings
should be the same). Transducers attached to anchors 1 and 2 give the move-
ment of these anchors relative to the reference head. The movement of anchors
1 and 2 with respect to each other and to the rock can be calculated by com-
bining the readings from transducers as required.
78
70
60 Sensor 1
Sensor 2
Sensor 3
50
Volumetric water content (Θ) (%)
Sensor 4
Θ = -439.501 + 10.507 Output Sensor 5
40 Sensor 6
Sensor 7
Sensor 8
30 Sensor 9
Θ = -4.242 + 0.798 Output Sensor 10

Sensor 11
20
Sensor 12
Sensor 13
Sensor 14
10
Sensor 15
5 10 15 20 25 30 35 40 45 50
Output (%)
Figure 4.9: Calibration of soil water sensors
4.3.8 Weather station

A weather station (all sensors supplied by Campbell Scientific Inc.) was
installed at the site to record the climatic conditions, with readings being
recorded at 10 minute intervals. The weather station consisted of;
• 1 tipping bucket rain gauge (model CS700) with a measuring range of 0

to 700 mm/hr at a resolution of 0.254 mm
• 1 anemometer (model 03101) to measure wind speed with a measurement

range of 0 to 50 m/s at a resolution of 0.5 m/s
• 1 pyranometer (model LI200X) to measure solar radiation with a mea-

surement range of 400 to 1100 nm at a resolution of 0.2 kW m−2 mV −1
• 1 temperature and relative humidity sensor (model HMP50) with a rela-

tive humidity measurement range of 0 to 98% RH at a resolution of ±3%
(0-90% RH); ±5% (90-98% RH) and temperature measurement range of
-25 °C to +60 °C at a resolution of < ±0.4% between 0 °C and +40 °C
The components of the weather station were tested in the laboratory for their
responses and the accuracy of measurements.
Figure 4.10: Rod-extensometer and the location of each installed anchor
4.3.9 Neutron probe

In addition to the soil water content sensors installed in the soil surrounding
the pipe, soil water content was measured using a neutron soil water gauge
(neutron probe) (503DR Hydroprobe). The neutron probe consists of a probe
containing essentially a fast neutron source (high-energy) and a slow thermal
neutron detector, a pulse counter (”ratescaler”), a cable connecting the two,
and a transport shield. Measurements are taken by lowering the probe to each
measurement depth by means of the cable via an access tube. A depth indicator
and clamp mechanism ensure consistency of depth between measurements, see
Figure 4.11.
The readings obtained from the neutron probe are affected by soil water
content, soil elemental composition, soil density, and proximity of the probe to
the water table and soil surface (Dickey, 1990). For these reasons site specific
calibration was undertaken. The results of field calibration are shown in Figure
4.12. Further detail on the calibration process can be found in Rajeev et al.
(2011).
4.3.10 Data acquisition system

All sensors, with the exception of the weather station, were connected to
a CR1000 datalogger and its peripherals, four AM16/32 multiplexers, one
AM25T multiplexer to suit thermocouples, AVWI vibrating wire interface for
80
Figure 4.11: Schematic of neutron probe
earth pressure cells, and one CE4 current excitation module for matric suction
sensors (Campbell Scientific Inc.). The CR1000 datalogger and peripherals
were mounted on a wooden board prior to field work to allow laboratory test-
ing of the data logging program and all sensors.
The CR1000 datalogger was programmed using a customised logging pro-
gram (written in CRBasic and original code provided by Campbell Scientific
Inc.). LoggerNet© software was used to communicate between the computer
and the datalogging system. The communication of each sensor with the log-
ging system was checked by performing benchmarking tests. The logging pro-
gram provided by Campbell Scientific Inc. was updated to incorporate cal-
ibration factors provided by the manufacturer or obtained from laboratory
calibration tests. Recorded data were saved to a 2GB industrial grade com-
pact flash card after each sensor cycle. Sensor cycles were conducted at 10
minute intervals.
The weather station sensors were connected to a CR800 datalogger (Camp-
bell Scientific Inc.), datalogger was programmed using a customised logging
program (written in CRBasic and original code provided by Campbell Scien-
tific Inc.). LoggerNet© software was used to communicate directly between the
computer and the datalogging system. The weather station also followed a 10
minute sensor cycle during which all weather station sensors were interrogated
and the output stored in the internal memory of the CR800.
Data were collected from the dataloggers on a monthly basis. Data recorded
by the CR1000 were collected by swapping the existing compact flash card
Figure 4.12: Calibration of neutron probe
with an empty one. Data recorded by the CR800 are collected via direct
serial connection between the datalogger and a laptop. In the field the data
acquisition system was housed in a meta cabinet and connected to mains power.
Figure 4.13 show the datalogging system after the connection of all sensors.
Figure 4.13a shows the laptop connected to the CR800 during data collection.
82
(a) Data acquisition system
Fan Thermocouple Multiplexer

control
Earth pressure sensor

Vibrating Wire multiplexer
Interface
Terminal blocks for sensor connection
Main Datalogger
(CR1000)
Weather Station
Power
Datalogger
(CR800)
outlet
Strain gauge
multiplexer
Power
outlet
Suction sensor
multiplexer
Fuse box and
safety switch
Moisture content
multiplexer
(b) Schematic
Figure 4.13: The data acquisition system installed at the site

4.4 Field work
Figure 4.14 shows the site plan with the locations of instrumentation pits and
the location of the bore hole for the installation of the rod extensometer. Pipe
and soil instrumentation was undertaken in three primary locations, designated
as Pit 1, Pit 2 and Pit 3. Each instrumentation pit contained two parts; part
A, for the instrumentation of the water pipe, excavated to 1.3 m below ground
surface, and part B, for the instrumentation of the surrounding soil, excavated
to 2.5 m below ground surface. As shown in the Figure 4.14, Pit 1 is located
beneath the driveway; Pit 2 is 3.65 m right of the driveway (5 m right of the
centre of Pit 1) and Pit 3 is 14.25 m right of the driveway (15.6 m right of the
centre of Pit 1). A small additional instrumentation pit, located 3.4 m away
from the driveway from from Pit 3, was also excavated for the installation of
pipe water pressure and temperature gauges. Figure 4.16 shows the vertical
cross-section indicating the location of soil sensors.
It was not possible to locate the joints in the water pipe prior to instrumen-
tation, as excavating along the pipe to locate joints would result in excessive
disturbance to the soil and that CCTV of the pipe interior posed too high a
risk to public health as it could possibly result in the contamination of the
potable water supply (the only CCTV cameras available were used to inspect
gravity sewers). Hence, the locations of sensors were judged on the basis of
intuitive assessment of the likely pipe behaviour by experienced field operators
from CWW and the typical length of pipe segments. The location of each
instrumentation pit was selected to monitor the strain of the pipe assuming
that it may be behaving like a fixed ended beam between the two driveways.
Pit 1 was at the end of the ”beam”, Pit 3 was located near the middle of the
”beam” (the location was shifted marginally towards Pit 1 due to a service
pipe connection) and Pit 2 was at approximately one third of the distance
between Pit 1 and Pit 3.
Figure 4.15 shows the detailed locations and labelling of the three sets
of biaxial strain gauges installed on the pipe. The location of the joints as
shown in Figure 4.15, is based on pipe joints being 6 m apart and the known
location of a joint found under the driveway during the excavation of Pit 1.
Figure 4.17 shows the detailed plan of a typical instrumentation pit; note that
matric suction, soil water sensors and thermocouples were installed above and
below the pipe by drilling horizontally through the soil from part B of each
instrumentation pit.
Property A Property B
15.6 m 15.2 m
1.1 m 0.4 m 1.3 m
3.16 m
1.58 m
2.4 m
Footpath
2.7 m 6.2 m 3.4 m 5.64 m
1.2 m
2.5 m
4.8 m
6.55 m Driveway
1 2 3.0 m 3
2.5 m
2.5 m
2.06 m 1.5 m
0.6 m
3.65 m 4.95 m
5.0 m 10.6 m
Road
23.8 m
1.0 m x 1.25 m x 1.0 m Instrumentation pit - part A Data logger position Circumferential Failure 2001 100 mm CI Water Pipe
Borehole Ø 75 mm Power pole 100 mm Gas Pipe
1.5 m x 1.5 m x 2.25 m Instrumentation pit - part B 1 Instrumentation pit number Sm all tree Storm Water Pipe
Service Connection - Water
1.0 m x 0.5 m x 0.9 m Instrumentation pit - pipe tapping Service Connection - Gas
84 Figure 4.14: A detailed plan of the Altona North instrumentation site
0.5 m 5.0 m 10.6 m 1.0 m
Earth pressure cell 1 Earth pressure cell 2
Strain 1L & 1C Strain 5L & 5C Strain 9L & 9C
Strain 4L & 4C Strain 2L & 2C Strain 8L & 8C Strain 6L & 6C Strain 12L & 12C Strain 10L & 10C
CHAPTER 4. FIELD INSTRUMENTATION
Strain 3L & 3C Strain 7L & 7C Strain 11L & 11C
Pit 1 Pit 2 Pit 3
Figure 4.15: Vertical cross-section and labelling of sensors in instrumentation pits - part A
85
Pit 1 Pit 2 Pit 3 Pit 3 - away from pipe
0.3 m
0.55 m
0.75 m
1.0 m
1.75 m
1.5 m
2.0 m
0.5 m 0.85 m 0.5 m 0.85 m 0.5 m 0.85 m 0.85 m 0.5 m

5.0 m 10.6 m
Thermocouple Sensors numbered from ground surface downward
Pit 1 contains sensors labelled 1 - 4;
Suction sensor Pit 2 contains sensors labelled 5 - 8;
Pit 3 contains sensors labelled 9 - 12
Soil moisture sensor Pit 3 also contains sensors 13 - 15 away from pipe
Figure 4.16: Vertical cross-section labelling of sensors in instrumentation pits - part B
86
1000
Pit part A
700
500
550
1500
Horizontal holes for

Pit part B sensor installation
Earth pressure cell
beneath pipe (pit 2 & 3 only)
1500
Strain gauges
100 mm CI Water pipe
Only in pit 3
Figure 4.17: Schematic of a typical instrumentation pit (plan view)
4.4.1 Excavation
The first task of the field instrumentation was to locate all buried pipes and ser-
vice connections on site to avoid damage. The location of each instrumentation
pit was then marked with spray paint (Figure 4.18(a)). Each instrumentation
pit was then excavated using an excavator. Where the excavation was close to
the pipe, the soil was manually removed using shovels and crowbars. Excava-
tion of Pit 3 can be seen in Figure 4.18(b). Once part B of each instrumentation
pit was excavated shoring was established to protect against collapse (Occu-
pational Health and Safety requirement, no shoring was required for part A of
each instrumentation pit) and the sensor installation was started. Prior to the
excavation of Pit 1 a section of the driveway was first cut out using a concrete
saw. Figure 4.18(c) shows a view of the site during instrumentation.
4.4.2 Sensor installation (7th - 14th of January 2008)

4.4.2.1 Strain gauges
As stated above part A of Pits 1, 2 and 3 exposed the water pipe for strain
gauging and installation of an earth pressure cell underneath the pipe in Pits
2 and 3. In total, twelve gauges, three sets of four biaxial strain gauges were
88
(a) Pit markup (b) Pit excavation
(c) Site work
Figure 4.18: Excavation for the instrumentation
installed. Each biaxial gauge consisted of two gauges: one gauge oriented along
the longitudinal axis of the pipe to measure the longitudinal strain and the
other oriented perpendicular to the first gauge to measure the circumferential
strain.
Installation of strain gauges on CI pipes is not straight forward, particularly
when the pipe is corroded and expected to be subject to future corrosion. As
it was crucial that the strain gauges remained viable for as long as possible,
to maximise the data collected, it was decided to commission Fortburn Pty.
Ltd., a company that specialises in strain gauging steel structures in harsh
environments, to install the gauges. Prior to strain gauging, the pipe surface
was initially cleaned using a sanding disk to remove dirt and graphite. The
pipe surface was then further cleaned with alcohol. The strain gauges were
attached to the pipe with specialised adhesive (Figure 4.20(a)). Curing of the
adhesive was accelerated using heated vacuum pads.
The use of the heated vacuum pads to accelerate adhesive curing required
that the original strain gauge wires were cut to ensure a good seal. After
curing the strain gauges were connected to 25 m lengths of durable shielded
data cable.
Once the cables had been connected, three layers of waterproofing mate-
rial (SEMKIT® ) were applied over the strain gauges, cable connections and
surrounding pipe to protect against moisture and inhibit corrosion. The first
layer of SEMKIT® was applied after the adhesive had cured; the subsequent
layers were applied after 24 and 48 hours from the application of the first
layer (Figure 4.20(b)). The strain gauge readings recorded after completion of
backfilling (12.30pm on the 12th of January, 2008) were taken as the reference
values for all subsequent results.
(a) Attaching strain gauges to pipe (b) Pipe after application of SEMKIT®
Figure 4.19: Strain gauging on the pipe
4.4.2.2 Earth pressure cells
Two earth pressure cells were placed approximately 50 mm below the invert
of the pipe in Pits 2 and 3. The earth pressure cells were installed by digging
a small hole in the wall of the pit beneath the pipe and preparing a flat base
using a thin layer of sand to ensure uniform contact between the pressure cell
and the soil (Figure 4.20a). In addition, a 15 mm thick, 200 mm diameter steel
plate was placed on the top of each earth pressure cell to ensure that the soil
pressure was uniformly distributed on the pressure cell (Figure 4.20b). The
hole was backfilled, using the removed soil, and compacted manually.
90
(a) Placement of earth pressure cell (b) Steel plate on earth pressure cell
Figure 4.20: Installation of an earth pressure cell
4.4.2.3 Soil monitoring sensors
The soil monitoring sensors; matric suction sensors, soil water content sensors
and thermocouples, were installed at four different levels above and below the
pipe from part B of Pit 1, 2 and 3. A fourth set of sensors were installed
at three levels in the soil away from the pipe and close to the road from
Pit 3. These sensors were installed to monitor the soil away from the pipe
for comparison with the measurements made at the pipe profile. The soil
monitoring sensors were installed by horizontally drilling holes with a hand
auger to allow installation of each sensor above or below the pipe (Figure
4.21). Good contact with the soil sensor is required for all three sensor types,
due to differences in the design of each sensor the installation method for each
sensor varied slightly.
Prior to installation of each thermocouple, a small cavity was prepared at
the end of the hole using a flat head screw driver. The thermocouple was
then inserted into this cavity and the surrounding soil compressed prior to
backfilling, Figure 4.22. Before installation of the matric suction sensors, a
piece of wooden dowel was pushed into the soil at the end of the hole to
create a small cavity slightly smaller than the ceramic head of the sensor. The
sensor was then inserted into the cavity, enlarging the cavity and ensuring good
contact with the soil. Figure 4.23 shows the installation of a soil water sensor
using a thin metal tube. Installation of the soil water sensor was simpler than
other sensors as no further preparation of the drilled holes was required. The
four pins on the head of the sensor were easily pushed into the soil. Figure 4.24
shows the installation of a soil water sensor using a short length of electrical
Figure 4.21: Horizontal drilling in preparation of sensor installation
conduit. The soil water sensors were installed horizontally to minimise the
measurement error.
4.4.2.4 Pipe water pressure and temperature gauges
Pipe water temperature and pressure gauges were installed using a custom
built T-piece and a pipe tapping connection. Figure 4.25 shows the gauges
after installation. The gauges were then enclosed in a bottomless plastic box
with a removable top plate. The top plate was level with the ground surface
and provides access to the gauges if required.
4.4.2.5 Wiring
After the installation of the sensors was completed, the wires were loosely
pinned to the wall of the pits to prevent damage during backfilling. The wires
were then laid out to their full length and cable tied together in preparation
for being sent through conduit into the instrumentation cabinet (Figure 4.26).
Once into the instrumentation cabinet, the wires were cut the length and
connected to the data acquisition system. The operation of all sensors was
checked prior to backfilling.
92
Figure 4.22: Installation of a thermocouple
Figure 4.23: Installation of a suction sensor
4.4.3 Backfilling
All pits were backfilled using the excavated material and compacted to a den-
sity close to the initial density (the same amount of soil excavated from each
pit was used to backfill - however no in-situ density measurements were con-
ducted). Backfilling was undertaken by layers up to the ground level (Figure
4.27(a)). Each layer was sprayed with water (Figure 4.27(b)) before compact-
ing with a vibrating plate compacter (Figure 4.27(c)). Where compaction was
close to the pipe and sensor wires the soil was compacted manually. Unfor-
tunately, a large amount of water (more than necessary for compaction) was
poured in to Pit 3 to wet the bottom soil layers prior to compaction of a layer
above them, this should be noted when viewing analysis of sensor information.
Figure 4.24: Installation of a soil water sensors
The top 30 cm of the Pit 1 (under the driveway) was backfilled with crushed
rock and a temporarily driveway created using a bitumen layer. A new drive-
way was installed several weeks after the field work was completed. The grassed
area of the site was restored after the instrumentation. Additional topsoil was
used as required, grass seeds were planted and watered by CWW. The amount
of water used to support grass growth is not known. Figure 4.27(d) shows the
instrumentation site after restoration was complete.
4.4.4 Supplementary sensor installations

4.4.4.1 Weather station (19th of February 2008)
The weather station was attached to the end of a galvanised steel pipe which
in turn was attached to the instrumentation cabinet. After installation, the
weather station was located 4.5 m above ground level. The weather station was
not located at a standard height due to site restrictions. The rain gauge was
levelled after installation to ensure accurate measurement. Sensor wire were
run through the galvanised steel pipe and directly into the instrumentation
cabinet where they were connected to the CR800 datalogger. The weather
station after installation and its components are shown in Figure 4.28.
4.4.4.2 Rod extensometer (4th of March 2008)
The rod extensometer was installed in a 75 mm diameter hole, bored to a

depth of 3.4 m, 1.4 m into the rock. The rod extensometer was inserted into
the hole such that anchors 3 and 4, which are at 2.0 m and 3.0 m from the
reference head, were situated in the rock. The section of extensometer within
94
Figure 4.25: Pipe water temperature and pressure gauges
the rock was grouted with cement, and anchors 3 and 4 inflated for maximum
contact with the cement.
Once the cement grout had hardened, anchors 3 and 4 were fixed to the rock
so that the movement of the reference head with respect to the rock could be
measured. The remainder of the hole was filled with a weak bentonite-cement
slurry and anchors 1 and 2, which are at 0.5 and 1.0 m from the reference
head, were inflated to embed into the soil. The reference head was located
0.4 m below the ground surface and covered with a wooden panel and top
soil. The wiring from the rod extensometer was run through small diameter
electrical conduit, buried approximately 100 mm below the ground surface.
The wiring entered the instrumentation cabinet via 90°conduit bends in the
instrumentation cabinet base.
4.4.4.3 Neutron probe (3rd of February 2009)
The access tube material selected for installation was aluminium as this mate-
rial has the least affect on readings compared to other available materials such
as stainless steel or brass. The access tube installed had an outside diameter
of 41.25 mm and wall thickness of 1.6 mm. The bottom of the tube was closed
with a tapered aluminium plug. The top of the access tube was closed with a
50 mm rubber bung to prevent water intrusion.
A hand soil auger was used to prepare the hole for access tube insertion.
Figure 4.26: Wiring after installation of sensors
The depth of the hole was greater than the length of the access tube to ensure
that the tube could be inserted to the maximum possible depth. The access
tube for the neutron probe was installed to a depth of 1.5 m. Insertion of
the access tube required use of a wooden block and mallet as the diameter
of the hole was slightly less than the access tube to ensure a tight fit. A
portion of soil around the opening of the access tube was removed and a steel
box installed level with the ground surface. Insertion of the access tube and
the fully installed access tube are shown in Figure 4.29(a) and Figure 4.29(b)
respectively. Further detail on the installation of the neutron probe and raw
data analysis is given in Rajeev et al. (2011).
4.4.5 Data Acquisition Timetable

Due to the need for supplementary sensor installation data records from all
sensors did not begin at the same time. Table 4.3 shows the timetable of when
the logging of each sensor began.
96
(a) Soil layer prior to compaction (b) Adding water to soil layer
(c) Vibrating plate compaction (d) Site after restoration
Figure 4.27: Backfilling and restoration of the site
Table 4.3: Data acquisition timetable

Sensor type Start date Start time Recording interval
th
Strain Gauges 12 January, 2008 12:30 pm 10 Minutes
Thermocouples 12th January, 2008 12:30 pm 10 Minutes
Water content
12th January, 2008 12:30 pm 10 Minutes
Sensor
Matric Suction
Sensor
Pipe Water
Pressure Gauge
Pipe Water
Temperature 12th January, 2008 12:30 pm 10 Minutes
Gauge
Weather Station 19th February, 2008 1:30 pm 10 Minutes
th
Rod Extensometer 4 March, 2008 1:30 pm 10 Minutes
Neutron probe
3rd of February 2009 N/A ≈ 1 Month
(Manual)
Figure 4.28: Weather station after installation at the site
(a) Insertion of access tube into augered hole (b) Access tube after completed installation
Figure 4.29: Installation of access tube for neutron probe

98
4.5 Investigation of field data

This section uses pipe stress data from the analysis conducted on the raw data
collected from the field instrumentation site. The data shown here was col-
lected from the times and dates detailed in Table 4.3 until the 17th of March
2010. The calculation of pipe stress accounted measured pipe axial and hoop
strains, internal pressure, burial depth, soil weight, pipe inside and outside
diameter, Young’s modulus, Poisson’s ratio, apparent thermal expansion co-
efficient and measured apparent thermal expansion coefficient. The analysis
of the raw data to calculate pipe stress from strain gauge data does not form
part of this thesis, full details on that analysis can be found in Chan (InPrep).
Selected results from individual sensors is given in Appendix B. The investiga-
tion of the pipe stress data in reference to the hypothesis proposed in Chapter
3 is solely a part of this thesis.
Chapter 3 proposed the hypothesis that soil shrinkage was the cause of the
peak in the number of pipe failures observed in mid to late summer. Specifically
this peak was the result of increase in circumferential failures. Circumferential
failures result from longitudinal stress (Rajani et al., 1996). In the case of soil
shrinkage the longitudinal stress would be a flexural stress resulting from the
bending of the pipe. Analysis of the stresses in the pipe over the course of the
instrumentation would reveal if these stress conditions developed. Additionally
correlation of the changes in stress with soil water content would confirm if
these stresses occur due to soil shrinkage as soil water content decreases.
4.5.1 Pipe stress

Based on the hypothesis proposed in Chapter 3 it is expected that pipe stress
will increase in mid to late summer and then reduce over winter being a min-
imum in spring before increasing again. A schematic representation of the
expected change in pipe stress and soil water content are shown in Figure
4.30. The shape of the curves shown is based on the shape of the intra-year
failure rate shown in Figure 3.17 and are not to scale.
Pipe stress due to soil drying is expected to increase as the pipe bends
downward toward the centre of the nature strip. The nature strip is expected
to dry faster as the distance from the driveway increases up to some maximum
rate. Pipe stress is expected to decrease as soil water content increases due to
rainfall.
Figure 4.31 shows the stress developed in the pipe since instrumentation
Summer Autumn Winter Spring Summer Autumn
Increasing value
Decreasing value
Expected pipe stress trend Expected soil water content trend
Figure 4.30: Expected trends in pipe stress and soil water content
overlayed with the expected trend from Figure 4.30 (it should be noted that
the values shown are not absolute stress, rather the change in stress referenced
back to the initial reading). The data shown for pipe stress has been limited to
the dates shown in Table 4.4. The data after these dates has been quarantined
from this analysis as it is believed to be erroneous, possibly due to delamination
of strain gauges. Data were deemed to be erroneous where rapid changes of
significant magnitude were observed.
Table 4.4: Investigated data timetable
Pit number Start date End date Recording window

th st
1 12 January, 2008 31 July, 2009 567 days
2 12th January, 2008 th
28 February, 2009 414 days
3 12th January, 2008 31st August, 2009 598 days
It can be seen that axial pipe stress varies over the course of the calender
year, peaking in mid to late summer as expected. Following this peak stress
reduces, also as expected. However the second peak expected in mid to late
summer 2008-09 was not clearly observed, if at all. There is some indication
100
Average Axial Stress (MPa)

Autumn Winter Spring Summer Autumn Winter
01/01/2008 01/03/2008 01/06/2008 01/09/2008 01/12/2008 01/03/2009 01/06/2009 01/09/2009

Expected Trend Pit 1 Pit 2 Pit 3
Figure 4.31: Average axial stress in pipe
of such a peak in the data for Pit 2 at the expected time, but no such peak
is observed for Pit 3 and data for Pit 1 indicated a peak occurring in late
spring early summer. These data cannot be said to support or contradict
the hypothesis proposed in Chapter 3; further analysis of the data was thus
required.
Further analysis were undertaken to determine the flexural stresses devel-
oped in the pipe in each instrumentation pit. Flexural stress analysis will also
indicate if pipe bending is occurring and the direction of that bending. For
the purpose of this analysis, flexural stress is defined as positive when the
crown is in tension relative to the invert at that location. The flexural stresses
developed in the pipe are shown in Figure 4.32 with 30 day moving average
trendlines, these plots are overlayed with the expected trend from Figure 4.30
(again it should be noted that this plot does not show absolute stress, rather
the change in stress referenced back to the initial reading).
The data shown in Figure 4.32 is consistent with the expected trend. It can
be seen that flexural stress increases in late summer to early autumn, this is
Flexural Stress (MPa)
01/012008 01/03/2008 01/06/2008 01/09/2008 01/12/2008 01/03/2009 01/06/2009 01/09/2009
Expected Trend Pit 1 Pit 2 Pit 3

Pit 1 trendline Pit 2 trendline Pit 3 trendline
Figure 4.32: Flexural stress in pipe
slightly later than expected but remains consistent with historical data, both
subsequent to instrumentation and the following year. The flexural stress is
seen to reduce over winter and be a minimum during spring.
In addition to being consistent with the expected trend, these results con-
firm an increase in longitudinal stresses as a result of flexure. The data suggests
that the pipe bends downwards in summer and upwards in winter. These data
support the hypothesis proposed in Chapter 3. The final step now required
to confirm the hypothesis is to show that this bending behaviour occurs as a
result of reducing soil water content.
4.5.2 Correlation of flexural stress with soil shrinkage

To confirm the final part of the hypothesis proposed in Chapter 3 the flexure
observed in the pipe must occur as the result of soil shrinkage due to soil water
content reduction. This was determined by comparing the pipe flexural stress
with observations of soil height change obtained from the rod extensometer.
102
The rod extensometer measured vertical soil movement with reference to

the rock located at a depth of 2 m. Rod extensometer Anchor 1 measured this
movement at a depth of 900 mm. This is the soil layer beneath the pipe. Figure
4.33 shows the vertical displacement of Anchor 1 and the average flexural stress
developed in the pipe.
It can be seen that as the height of the soil layer increased, the flexural
stress in the pipe decreased. As the height of the soil layer decreased, the
flexural stress in the pipe increased. The timing of these observations is such
that pipe flexural stress is seen to increase during mid to late summer in both
2008 and 2009, and decrease in between. This provides strong evidence that
soil shrinkage is the cause of the flexural stress.
Change in soil height (mm)

Flexural stress (MPa)
01/03/2008 01/06/2008 01/09/2008 01/12/2008 01/03/2009 01/06/2009 01/09/2009
Average flexural stress Anchor 1 (900 mm)
Figure 4.33: Change in height of soil beneath pipe (measure by rod exten-
someter) with average pipe stress
4.5.3 Pipe-soil behaviour

The results of the investigation provides strong evidence that the hypothesis
proposed in Chapter 3. Using this hypothesis, it is possible to anticipate the
behaviour of the pipe in response to changes in climatic condition over the

course of the year. Figure 4.34 shows a prediction of pipe movement as a
result of changes in soil water content in summer and winter. In summer,
the soil water content decreases resulting in soil shrinkage causing the pipe to
bend downward; in winter, the pipe displacement is reversed as the soil water
content increases causing the soil to swell.
Nature strip
Driveway Driveway
Soil shrinkage
A Tension Tension C
Compression B Compression
Summer Time
Compression Compression
Tension Tension
Soil swelling
Winter Time
Figure 4.34: Predicted vertical pipe movement in winter and summer based
on the stress analysis
4.6 Conclusions
The field study was undertaken on an in-service CI reticulation pipe with a
100 mm nominal diameter buried in a soil with high shrink/swell potential.
Analysis of the results showed that between mid summer to early autumn
positive flexural stress (positive longitudinal stress at the pipe crown, negative
at the invert) developed in the pipe. Positive flexural stress indicates that the
the pipe was bending downward towards the centre of the nature strip relative
to the driveway. The development of positive flexural stress was the direct
result of soil shrinkage due to soil drying at this time. Subsequent wetting of
the soil resulted in a reversal of the flexure developed during drying, eventually
leading to negative flexural stress developing in the pipe.
The detailed field study presented in this chapter has given strong support
for the hypothesis presented in Chapter 3 that the increase in pipe failure num-
bers during mid summer to early autumn occurs as a result of soil shrinkage.
Chapter 5
Mechanical properties of
Australian Cast Iron
reticulation pipes
5.1 Introduction
This chapter is the first of four which fulfills Stage three of the research pre-
sented in this thesis. The aim of Stage three is to develop a modelling approach
to represent the development of pipe stresses resulting from soil shrinkage. This
chapter investigates the mechanical properties of Australian pipes. Chapter 6
presents a novel approach for model development, which is applied in Chap-
ter 7 to develop a model for the volume change behaviour of soil in response
to changes in soil water content. Chapter 8 combines the results from these
investigations to produce the final model.
This chapter focuses on the mechanical properties of Australian cast iron
(CI) pipe. CI was identified as belonging to the cohort of pipes most affected
by seasonal climatic changes, based on the results of the analyses presented in
Chapter 3. Additionally, a CI pipe was the focus of the field study detailed in
Chapter 4.
The mechanical properties of CI pipes has been the subject of a compre-
hensive investigation in North America. Less comprehensive studies from the
United Kingdom and Japan have also been conducted. For this reason the
work detailed in this chapter was conducted as a scoping study to allow a
comparison with these data.
Mechanical testing was conducted using samples exhumed from Melbourne’s
water and natural gas reticulation pipe networks. Samples of natural gas retic-
105
106
ulation pipe were used due to a limitation in the availability of water pipe
samples at the time of this work. Testing was undertaken to determine the
tensile strength, secant modulus, fracture toughness and fatigue behaviour.
These results were then be compared to literature values.
Fatigue testing results are also used to investigate the likelihood of fatigue
failure occurring in reticulation pipes using data collected from the field study
(Chapter 4)
Tensile strength and fracture toughness results were used to investigate the
net section collapse and crack growth pipe failure modes for pipe exhibiting a
range of pit depths due to graphitisation.
Section 5.3 summarises the existing knowledge on the mechanical properties
of CI reticulation pipes. Section 5.5 details the specimens used for mechanical
assessments. Section 5.6 presents the results of the tensile strength, modulus
and fracture toughness assessments, and compares them with those presented
in the literature. Section 5.7 investigates the effect of fatigue on pipe failure.
Section 5.8 models a pipe segment subjected to bending to determine the most
likely failure mode.
5.2 Background
The most common pipe type in the Melbourne water reticulation networks is
DN100 (100 mm nominal diameter) grey CI (henceforth referred to as CI).
This pipe type is also the oldest currently in-service and is experiencing the
highest failure rates of all pipe types. The average failure rate of CI pipe over
twelve years (1996 - 2007 inclusive) exceeded 72 failures/100 km/year. This
contrasts strongly to the failure rate for the remainder of the network over the
same period, which was 20 failures/100 km/year.
A major requirement for understanding the performance of a pipe network
is knowledge of the factors which contribute to its failure, as described in Chap-
ter 2. Failure occurs when the forces applied to a pipe exceed its structural
capacity; therefore knowledge of the mechanical properties of pipes is crucial to
understanding pipe performance. Previous researchers have conducted testing
to determine the mechanical properties of CI pipes on samples exhumed from
water reticulation networks (Atkinson et al., 2002, Makar et al., 2005, Rajani
et al., 2000, Seica and Packer, 2004, Yamamoto et al., 1983). These studies
were conducted on pipes from the UK, Canada, the USA and Japan. However,
there are no studies in the literature on the CI pipes from Australian water
CHAPTER 5. MECHANICAL PROPERTIES CAST IRON PIPES 107
reticulation networks. In light of the variation observed in the properties of CI

pipes reported in the above studies, it is necessary to quantify the properties
of Australian CI and to compare the results with the results of international
studies. To this end, an array of laboratory experiments was conducted on CI
pipe samples to determine their mechanical properties. The pipe samples were
exhumed during the course of repair works. The detailed scheme of labora-
tory experiments performed on these pipe samples is discussed in the following
section.
5.3 Previous work

Previous research into the mechanical properties of CI water pipes often aggre-
gated results from pipe manufactured using both pit and spun cast methods.
Where results have been disaggregated by casting method, it has been shown
that pit cast pipe has generally lower values for both tensile strength and frac-
ture toughness than spun cast pipe; however significant overlap can also be
seen as shown in Table 5.1. It can also be seen that whilst all researchers have
measured tensile strength, only two have investigated fracture toughness. The
values report by each author showed significant variation, particularly in the
case of tensile strength.
Table 5.1: Properties of uncorroded CI from previous studies
Fracture Toughness
√
Source Casting Method Age Range Tensile Strength (M P a)
(M P a m)
Kirkby (1977)† Pit & Spun Cast 19-77 years 150-170 N/A
Colin and Baker (1991)† Pit & Spun Cast Ex-service pipes 137-212 10.5-15.6
Atkinson et al. (2002) Not Specified Not Specified 170 - 221 13.6
Makar et al. (2005) Not Specified Not Specified 220-260 N/A
Rajani et al. (2000) Pit Cast 64-115 years 33-267 5.7-13.7
Rajani et al. (2000) Spun Cast 22-61 years 135-305 10.3-15.4
Seica and Packer (2004) Pit & Spun Cast 50-124 years 47-297 N/A
Yamamoto et al. (1983) Not Specified 22-79 years 104-167 N/A
Caproco Corrosion
Spun Cast 22 - 28 years 70 - 217 N/A
Prevention Ltd (1985)†
Ma and Yamada (1994)§ Spun Cast 21-32 years 40-320 N/A
Philadelphia Water
Not Specified 20-130 231‡ N/A
Department (1985)†
†
Source: Rajani et al. (2000)
§
Source: Seica and Packer (2004)
‡
Upper range tensile strength inferred √
from data (Rajani
√ et al., 2000)
1 M P a = 0.145 ksi; 1 M P a m = 0.920 ksi in
108
5.4 Material deterioration

As noted in Chapter 2 the failure of buried pipes occurs when the applied loads
exceed the structural capacity of the pipe. Failures occurring close to the time
of installation are attributed to manufacturing defects and installation issues
such as settlement. Failures occurring after this initial period are attributed to
the deterioration of the pipe structural capacity or increases in applied loads.
The mechanism of this deterioration is dependant on the pipe material, for
CI and other metallic pipe materials deteriorate predominantly as a result of
electro-chemical corrosion (Makar, 1999, Rajani and Kleiner, 2001).
In CI, electro-chemical corrosion results in the removal of iron from the pipe
wall, leaving the graphite matrix in place, a process known as graphitisation.
The extent of graphitisation is often difficult to determine by visual inspection
as the surface of the pipe can appear unchanged. To determine the extent of
corrosion using visual or mechanical methods, the graphite matrix must first
be removed; one method by which this can be done is by grit (or shot) blasting.
Due to the importance of graphitisation in the reduction of CI pipe capacity
the effect of graphitisation on CI tensile strength and secant modulus has been
included in this investigation.
5.5 Experimental Work
5.5.1 Sample Details

The samples used in this investigation were exhumed from the reticulation net-
works during routine repair works in 2007. The pipes from which the samples
were extracted ranged in age from 34 to 123 years. All samples obtained from
water reticulation networks were cement mortar lined either during manufac-
ture or in-situ. No lining was present on the samples obtained from the gas
reticulation network. The pipes were located in a variety of soil environments
and were all installed using native backfill.
5.5.2 Tensile strength and secant modulus

Dog bone specimens for determination of tensile strength and secant modulus
were machined using a CNC (computer numerical control) mill in accordance
with ASTM E8M-04 (ASTM International, 1999), see Figure 5.1. However,
the length of the grip sections were reduced from the required 75 mm to 60
110
mm to suit the testing machine. The specimens were machined along the
pipe axis at six equidistant locations around the circumference of the pipe
samples. Specimens which could not be machined to the required dimensions
were discarded.
Tensile strength was calculated as the tensile stress at failure (the maximum
tensile stress) using Equation 5.1; secant modulus was calculated from the
tensile strength and strain at failure using Equation 5.2; strain is calculated as
shown in Equation 5.3, see Figure 5.2. Tensile testing was conducted at 0.25
mm/min using an Instron material testing machine.
F
σ= (5.1)
A
where σ is the tensile stress, F is the tensile force and A is the cross-sectional
area of the test length.
σ
Es = (5.2)
ε
where Es is the secant modulus, σ is the tensile strength and ε is strain at
failure.
∆l
ε= (5.3)
l0
where ε is strain, ∆l is the change in sample test length and l0 is the initial
sample test length.
8 mm 60 mm 60 mm 60 mm
50 mm
20 mm
Ø 25 mm
Grip section Test length Grip section

Fillet section Fillet section
Figure 5.1: Tensile specimen dimensions
5.5.3 Fracture toughness

Fracture toughness testing was undertaken following ASTM E 399 - 06 (ASTM
International, 2006). However, Double Edge Notched Tension (DENT) test
specimens were used instead of the specified Single-Edge Notched Bend (SENB)
or Compact Tension (CT) specimens. By definition, the fracture toughness
values obtained from either specimen types should be the same, however test-
ing conducted by Rajani et al. (2000) indicated that results obtained using
Tensile
strength
us
ul
od
tm
an
c
Se
Figure 5.2: Tensile specimen dimensions
SENB specimens from pit and spun CI pipe samples produced fracture tough-
ness values up to 25% greater than those obtained using CT specimens. For
this reason the more conservative CT specimens were preferred. However, the
use of this specimen type was not possible due to size constraints in regards
to pipe wall thickness, therefore DENT test specimens were adopted in line
with Rajani et al. (2000), see Figure 5.3. DENT specimens were also used for
investigation of fatigue behaviour.
Specimens were machined along the pipe axis at locations around the pipe
circumference. The location of each specimen was chosen to minimise the
presence of graphitisation. Specimens which exhibited graphitisation across
the fracture surface after testing were discarded and have not been included
in the results reported. Fracture toughness was determined using the solution
from Janssen et al. (2004), shown in Equation 5.4.
 
2 3 4 

 d d d d
− − −

√

 1.122 1.122 0.820 + 3.768 3.040 

 W W W W 
KI = σ πd s

 d 


 1−2 

W
 
(5.4)
where KI is the applied stress intensity factor, σ is the stress, d is the notch
112
depth and W is the sample width. Sample stress is calculated for the sample
cross-section between the notches.
180 mm 8 mm
20 mm
10 mm
90°
Figure 5.3: DENT specimen dimensions
5.5.4 Fatigue
Fatigue testing was undertaken using fracture toughness DENT specimens.
Specimens were tested at a range of values of ∆KI (change in KI ). The max-
imum KI applied to any specimen was 60% of the fracture toughness deter-
mined for the sample from which that specimen was obtained. The minimum
KI applied was calculated as a proportion of the maximum KI . Values for
the minimum KI varied between 0.1 and 0.5. An example of these calcula-
tions is given in Table 5.2. All fatigue tests were undertaken for tension only
conditions.
Table 5.2: Example calculations for fatigue testing

Specimen Fracture Maximum √ KI Minimum √ KI
Sample ID
ID Toughness (M P a m) (M P a m)
FT1 10.9 N/A N/A
F1 N/A 10.9 × 0.6 = 6.54 6.54 × 0.5 = 3.27
Sample 1
F2 N/A 10.9 × 0.6 = 6.54 6.54 × 0.3 = 1.96
F3 N/A 10.9 × 0.6 = 6.54 6.54 × 0.1 = 0.65
5.6 Results and Discussion

5.6.1 Tensile strength
A total of 50 specimens were machined from 12 pipe samples and underwent
tensile testing. Of the 50 specimens tested, 20 were discarded due to problems
associated with testing, such as the specimen failing in the grip or fillet sec-
tions, this was usually associated with the presence of graphitisation in these
locations. Figure 5.4 shows a typical stress-strain curve as observed during

testing. The maximum strain values shown for each specimen were less than
expected from literature. The values in the literature indicated failure between
strain values of 0.0027 mm/mm and 0.0077 mm/mm, and between stress val-
ues of 133 MPa and 298 MPa (Makar and McDonald, 2007, Seica and Packer,
2004, 2006). However, the lower than expected values are understandable as
both specimens shown in Figure 5.4 contained graphite pits which reduced
their capacity.
A plot of tensile strength against age is shown in Figure 5.5. The casting
methods identified for each test specimen in this figure were inferred from the
pipe sample installation dates and the installation periods shown in Table 5.3.
The results varied widely during testing both between pipe samples and even
between specimens machined from the same pipe sample. The overall range
of tensile strength results was between 10.5 MPa and 249 MPa. Interestingly,
both results came from spun cast pipe samples. As was found by Seica and
Packer (2004) and Rajani et al. (2000) no correlation between age and tensile
strength could be inferred from the results. The large variation in these results
is most likely due to the graphitisation defects present in the test specimens.
This lack of correlation supports the statement in Chapter 2 that pipe age
alone is not a reliable indicator of pipe condition.
Figure 5.4: Stress-Strain curve for Specimens C2-T1 (Spun Cast) and S2-T6
(Pit Cast)
114
Table 5.3: Cast Iron installations periods (Scott, 1990)

Installation Period Pipe Type
< 1921 Horizontal pit CI made in Australia
>1921 <1929 Horizontal pit CI made in Australia
>1921 <1929 Imported spun CI
>= 1929 Spun CI made in Australia
>= 1981 Ductile iron
Figure 5.5: Uniaxial Tensile Strength of Grey Cast Iron against Sample Age
The effect of graphitisation on tensile strength was investigated by plot-

ting the ratio of maximum graphitisation depth to specimen thickness (a/t)
against tensile strength following the method used by Atkinson et al. (2002),
Figure 5.6. The maximum graphitisation depth was measured on the fracture
surface, using an Olympus optical microscope with digital camera and mea-
surement software calibrated against precision slip gauges. Fracture surfaces
were initially imaged and measured before grit blasting. The samples were
remeasured after grit blasting and the results compared. It was found that
accurate measurements were possible without grit blasting.
The coefficients of determination (R2 ) shown in Figure 5.6 indicates a rea-
sonable correlation between a/t and tensile strength, particularly in light of
the large variation of mechanical properties inherent in the material. A clear
difference can be seen in the properties of pit and spun cast pipe, in terms of
tensile strength at a/t values of zero and the rate of tensile strength reduction
as a/t increases.
The trend lines shown in Figure 5.6 indicate that pit cast pipe has a tensile
strength of 36 MPa lower than the tensile strength of spun cast pipe, when a/t
is equal to zero. The trend line for pit cast pipe suggests that some strength
is retained when a/t is equal to one. This confirms research by Atkinson et al.
(2002) who reported that some strength is retained when a/t is unity. This is
also seen in spun cast specimen, C3-T5, which had undergone graphitisation
through its thickness prior to testing. However, the spun cast specimen trend
line indicates that a complete loss of strength occurs when a/t = 0.92. This
result is believed to be an artefact of the insufficient number of test specimens
with high a/t ratios and not a true indication of material behaviour.
= 0.62
(a/t)
Figure 5.6: Correlation between graphitisation depth to thickness ratio and

tensile strength
Comparison of the results plotted in Figure 5.6 (Australian samples) with

results from tests on graphitised samples conducted in the UK (Atkinson et al.,
2002), Japan (Yamamoto et al., 1983) and Canada (Rajani et al., 2000) in-
dicate that there is a significant variation in the properties of the materials
tested. Trendlines provided by the sources of this data have been recalculated.
Data points obtained from these sources have not been included for clarity.
Figure 5.7 shows the line-of-best-fit trendlines for the data from of these coun-
tries. The data on which the trendlines are based have been separated by
casting method where this was available.
116
Graphitisation Depth / Thickness (a/t)
Figure 5.7: Comparison of results from UK, Japan, Canada and Australia
(only trendlines shown)
The trendlines shown in Figure 5.7 indicate that all materials (with the
exception of Australian spun CI as discussed above) showed a residual strength
even with graphitisation occurring through the thickness of the samples. It can
also be seen that there is a clear difference in the properties of each CI type
plotted, with regards to average un-graphitised tensile strength. It can also be
seen that most materials show similar rates of tensile strength reduction with
increasing graphitisation depth/specimen thickness ratio. Australian spun CI
did not show either of these properties, showing a complete loss of strength
at a graphitisation depth/thickness ratio less then 1.0 and reducing in tensile
strength at a higher rate than all other materials, almost double the rate
calculated for the Japanese data. Whilst there is clear variation in the results
from different sources, the results obtained for Australian CI are within the
bounds of the published data.
5.6.2 Secant modulus
The secant modulus of elasticity was calculated for each of the tensile test
specimens. The point of failure was defined as the point of maximum stress
applied to the specimen during testing. Figure 5.8 shows the secant moduli
determined during testing. The specimen age and casting method inferred
from installation date are also shown. Investigation of relationships between

secant modulus and tensile strength, and secant modulus and graphitisation
revealed little or no correlation.
Figure 5.8: Secant Modulus
5.6.3 Fracture toughness

A total of 27 fracture toughness specimens were machined for testing from five
pipe samples and underwent testing. Of these specimens, the results from eight
were discarded due to problems associated with testing, such as graphitisation
at the fracture surface or the specimen failing away from the notched section
(also associated with graphitisation). Table 5.4 shows the results from the
fracture toughness tests. The casting method for the specimens shown in Table
√
3 were inferred using average values for fracture toughness of 10.2 M P a m
√
and 13.2 M P a m reported for Canadian pit and spun CI pipes respectively
(Rajani et al., 2000).
Comparison of the casting method inferred from the data in Table 5.3 and
those shown in Table 5.4 revealed consistent results from all pipe samples
except pipe sample C5. The casting method for sample C5 based on fracture
toughness was pit casting, whilst the installation date indicated that the pipe
sample was spun cast. The fracture toughness results obtained for sample
C5 are very low in comparison to those reported for spun CI, in fact only
Table 5.4: Fracture toughness test results
Casting method inferred Fracture toughness
√ Average fracture
√
Sample ID Specimen ID Age
from fracture toughness (M P a m) toughness (M P a m)
C1 D1 87 Pit 9.7
C1 D2 Pit 10.9
10.6
C1 D3 Pit 11.9
C1 D4 Pit 10.0
C3 D1 86 Spun 12.2
12.9
C3 D2 13.6
C4 D1 56 Spun 12.1
C4 D2 Spun 12.9
C4 D3 Spun 11.8
12.9
C4 D4 Spun 13.1
C4 D5 Spun 13.0
C4 D6 Spun 14.3
C5 D1 46 Pit 11.2
C5 D2 Pit 11.7
C5 D3 Pit 10.1
10.2
C5 D4 Pit 8.5
C5 D5 Pit 9.8
C5 D6 Pit 9.7
S2 D1 123 Pit 9.6 9.6
118
two of the six values fall within the range of reported values shown in Table
1 for spun cast pipe. Measurement of the wall thickness of the pipe sample
showed negligible variation (<1 mm), indicating that the pipe was more likely
to have been spun cast as pit cast pipes often exhibit significant dimensional
variation (Rajani et al., 2000). This case highlights the variation in material
properties inherent to grey CI pipes; the specific reason for the deviation from
the expected result is however unknown.
The average fracture toughness values from this investigation, calculated
from the average results of each pipe sample for the pit and spun CI were
√ √
10.1 M P a m and 12.0 M P a m respectively. Note that sample C5 has been
excluded from this calculation due to inconsistency in the casting method
inferred from installation date and dimensional consistency, and the measured
fracture toughness.
5.6.4 Fatigue
Previous research into the behaviour of in-service pipes has indicated that the
applied loads vary over daily and yearly time-steps. The yearly variation in
load is believed to be the result of restrained thermal expansion/contraction
(Kuraoka et al., 1996) and the action of expansive soils (Gallage et al., 2009),
whilst the daily load variation is the result of water usage patterns affecting
internal pressure (Gallage et al., 2009).
Due to the presence of these varying loads, fatigue testing was undertaken
to determine the influence, if any, of fatigue behaviour on pipe failure. Fatigue
testing was conducted on specimens obtained from four gas reticulation pipe
samples, not water reticulation pipe as used in other testing. To determine
the applicability of using the gas reticulation pipe for this testing, fracture
toughness and tensile strength tests were conducted on specimens from each
pipe sample used. It was found that the properties of gas reticulation pipe
were comparable to those of water reticulation pipe.
The number of cycles to failure for each specimen is shown in Figure 5.9.
Unfortunately data on the rate of crack growth during testing, could not be
collected. These results are used in Section 5.7 to determine if fatigue failure
is likely to occur in reticulation mains.
120
ΔK (MPa√ m)
1,000 10,000 100,000 1,000,000 10,000,000

Cycles to Failure (N f )
Figure 5.9: Fatigue Test Results
5.7 Fatigue failure prediction

The high degree of scatter evident in the fatigue testing results shown in Figure
5.9 precludes its use for reliable prediction of failure. However, an indication
of the importance of fatigue on reticulation pipe failure can still be obtained.
The preferred data for prediction of fatigue lifetime using a fracture mechanics
approach is the crack growth rate for the applied conditions. Unfortunately
these data were not obtained during testing as described in Section 5.6.4.
However, it is possible to determine fatigue lifetime using a simplified approach.
The fatigue lifetime of a CI pipe was predicted solely on the basis of initial
conditions. This method is expected to over estimate the lifetime as it does
not explicitly account for the increase in KI due to crack growth.
In order to investigate the failure of CI pipe using a fracture toughness ap-
proach it is necessary to determine the geometry of a ”normal” graphitisation
pit. To achieve this an analysis was conducted on data presented by Rajani
et al. (2000) which provided dimensions for 146 graphitisation pits. For the
purpose of this analysis it was assumed that the ratio of pit width to pit depth
(W/D) remained constant regardless of pit depth. Based on this assumption,
the W/D ratios of each pit were fitted with a number of statistical distri-
butions. The log-normal distribution was found to have the best fit (lowest
standard error) and is shown in Figure 5.10. This distribution was then used
to determine the W/D ratio of a ”normal” graphitisation pit; the expected

W/D ratio was 3.5:1.
0.30
0.25
0.20
Probability
0.15
0.10
0.05
0.00
0 2 4 6 8 10 12 14
Graphitisation Pit Width/Depth
Figure 5.10: Log-normal distribution fitted for pit geometry
The total variation in pipe axial and flexural stresses determined from the
field study site (Chapter 4), were 38 MPa and 52 MPa respectively. Using
the instrumented pipe as a basis (spun CI, 120 mm outside diameter and 8.5
mm wall thickness), the expected time for fatigue failure was modelled. The
instrumented pipe was located in a nature strip and not subject to traffic
loading.
∆K was calculated, assuming a pit depth equal to half the pipe wall thick-
ness, using the solution by Chapuliot (2000) for a tube with an elliptical exter-
√
nal defect, Equation 5.5, and determined to be around 6.5 M P a m. Expected
life was calculated based on the sample which showed the highest rate of de-
crease of cycles to failure as ∆K increased (sample 1 shown in Figure 5.9). The
expected life of this pipe, assuming no other factors are acting (such as graphi-
tisation and change in applied loading), is around 28,000 years. As shown in
√
Figure 5.9 one specimen was tested at ∆K of around 6.5 M P a m; this sample
failed after 7,200 cycles, which in the modelled scenario is equivalent to 7,200
years. These results indicate that for reticulation pipes fatigue is not an impor-
tant consideration in the prediction of time to failure. A further investigation
which calculated fatigue lifetime over time based on a graphitisation rate of
0.1 mm/year found that failure (determined by either the predicted fatigue life
or full wall thickness graphitisation) was controlled by graphitisation rate.
122
" #
√
2 3
d d d
K I = σ 0 i 0 + σ 1 i1 + σ2 i2 + σ 3 i3 + σgb Fb πd (5.5)
t t t
where KI is the applied stress intensity factor, σ0 , σ1 , σ2 ,σ3 and σgb are stresses
associated with different load types, i0 , i1 , i2 , i3 and Fb , are geometric correc-
tion factor associated with these stresses respectively, d is the notch depth and
t is the pipe wall thickness. The value of the geometric correction factor are
dependant on both pit and pipe geometries, and are given in Chapuliot (2000).
5.8 Failure mode comparison

Atkinson et al. (2002) reported that both loss of section and fracture mechanics
approaches provide reasonable methods for predicting capacity reduction with
increasing pit depth. Based on this finding an analysis was conducted to
determine the most likely mode for pipe failure over a range of graphitisation
pit depths. However, the method for capacity reduction used by Atkinson et al.
(2002) was not followed. An alternate method, based on the loss of material
due to graphitisation, rather than a reduction in the material strength of CI
was used. More detail is given later in the section.
This section compares the uniformly distributed loads required to cause
the failure of a graphitised pit cast pipe due to net section collapse (i.e. tensile
failure due to exceeding tensile strength) and crack growth (i.e. fracture failure
due to exceeding the fracture toughness). Previous modelling of the response
of a buried pipe to loading by Chan (2008) found that the end conditions
applicable to buried pipes are in between the simple cases of pinned and fixed
ends. For this reason both extremes have been investigated here.
For the purposes of this analysis, a worst case scenario has been investigated
where the pipe span is unsupported, the graphitised area is located at the mid-
point of the span, at the base of the pipe. The depth of the graphitised area
was varied between 5% and 95% of the pipe wall thickness. The contribution of
the graphite structure retained after graphitisation is assumed to be negligible,
i.e. graphitised areas have no residual strength. Calculation parameters are
given in Table 4.
Failure resulting from net section collapse was deemed to have occurred
when the maximum tensile stress in the pipe was equal to or exceeds the tensile
strength given in Table 5.5. For the purpose of this analysis the reduction in
Table 5.5: Failure scenario calculation parameters

Casting Method Pit Cast
Tensile Strength 198 M P√ a
Fracture Toughness 10.1 M P a m
Outside Diameter 126 mm
Wall Thickness 8 mm
Span 6m
Pit Width/Depth Ratio 3.5 : 1
pipe structural capacity was not determined based on the trendlines shown
in Figure 5.6, as those trendlines indicate a reduction in the strength of CI
rather than the more realistic loss of material. In this analysis the capacity
of the CI material remains constant, but the pipe structural capacity reduces
due to a decrease in the pipe wall thickness. The reduction in wall thickness
was assumed to be constant around the circumference of the pipe, known as
general graphitisation. The decrease in the pipe wall thickness, results in a
decrease in the pipe’s second moment of area and a corresponding increase in
pipe stress for any given uniformly distributed load.
Failure resulting from crack growth was deemed to occur when the applied
stress intensity factor equalled or exceeded the average fracture toughness given
in Table 5.5. The applied stress intensity factor was calculated based on cir-
cumferentially aligned elliptical pit at external surface of a tube (Chapuliot,
2000). The geometry of the pit was assumed to be that of a ”normal” graphi-
tisation pit, the W/D ratio of which remains constant as pit depth increases
(as was assumed in Section 5.7).
The uniformly distributed loads required to cause failure are shown in Fig-
ure 5.11. It can be seen that for all graphitisation depths failure will occur as
the result of net section collapse. However, it should be noted that this assumes
that general graphitisation would occur at the same rate as pit graphitisation.
This assumption is not realistic as CI buried in soil is more likely to undergo
pitting graphitisation (Reynaud, 2010). Therefore, for any given pit depth the
comparative depth of general graphitisation could be substantially less.
A simplified method is used here here to relate pit graphitisation depth to
a comparable general graphitisation depth, based on the volume of material
lost. The volume of material lost due to both graphitisation types is assumed
to be equal, and general graphitisation occurs as a band equal in width to the
graphitisation pit. Using this assumption a pit depth/wall thickness ratio of
0.5 (4 mm pit depth, 14 mm pit diameter) equates to a general graphitisation
124
6000
Net section collapse - fixed-fixed
Net section collapse - pinned-pinned
Crack growth - fixed-fixed
5000 Crack growth - pinned-pinned
4000
UDL (N/m)
3000
2000
1000
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Graphitisation depth/Wall thickness
Figure 5.11: Failure Mode Comparison
depth/wall thickness ratio of 0.07 (0.6 mm general graphitisation depth, 14

mm general graphitisation band width). Using this comparison failure would
occur as the result of crack growth. Using this comparison method it can be
seen that a large pit depth equates to a small depth of general graphitisation.
Indicating that crack growth is the more likely failure mode at this pit depth.
Figure 5.12 shows a modified version of Figure 5.11 where graphitisation pit
depth is plotted with the comparable general graphitisation depth, determined
using the equivalent volume criteria. From Figure 5.12 it can be seen that
failure by net section collapse is likely to occur at smaller graphitisation depths,
up to a pit depth/wall thickness ratio of approximately 0.25. For larger pit
depths failure is more likely to occur as the result of crack growth.
General graphitisation depth/Wall thickness

0 0.003 0.012 0.027 0.048 0.074 0.107 0.147 0.192 0.2438 0.302
6000
5000
4000
UDL (N/m)
3000
2000
1000
Net section collapse - fixed-fixed
Net section collapse - pinned-pinned
Crack growth - fixed-fixed
Crack growth - pinned-pinned
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Graphitisation pit depth/Wall thickness
Figure 5.12: Failure Mode Comparison
5.9 Conclusions
This chapter has presented the mechanical property assessment of Australian
CI reticulation pipe. Results from all tests showed high variability between
the samples tested. The segregation of CI samples based on casting method
inferred from installation date reduced the variability in results. Comparison of
test results with results from a comprehensive investigation of North American
CI reticulation pipe found comparable results. For this reason further, more
comprehensive, testing was not undertaken.
The investigation of the fatigue failure as a possible cause of pipe failure
found that fatigue was unlikely to have a significant affect on pipe failure.
Fatigue crack growth was considered to occur at a rate less than material
126
deterioration due to graphitisation.

Modelling was undertaken to determine the most likely failure mode at
varying levels of graphitisation. Based on graphitisation depth alone net sec-
tion collapse was found to be the most likely failure mode. However, this as-
sumed that general and pit corrosion depth increased at the same rate. When
failure modes were compared on the basis of volume of material lost the failure
was most likely to occur by net section collapse for pit depth/wall thickness
ratios up to approximately 0.25. Above this value failure was more likely to
occur as the result of crack growth.
Chapter 6
A novel technique for model

development
6.1 Introduction
Experimental measurement of soil behaviour in response to changes in soil
water content allows soils to be characterised. However, in order for this
data to be used as part of a larger model, the response must be represented
mathematically. This chapter develops a technique for the creation of empirical
models, where characteristic features of the relationship to be modelled are
directly related to fitting parameters of the model. This direct relationship
will allow additional factors, which affect individual characteristic features,
to be incorporated into the existing model without the need to develop an
entirely new model. The technique presented also ensures that it is easy to
obtain the derivative of the model. The technique is applied in this chapter to
develop a model for the soil water characteristics curve (SWCC), a fundamental
relationship for the behaviour of water in soil.
Section 6.2 details the importance of the SWCC and the need for a tech-
nique to develop models where fitting parameters are directly related to char-
acteristic features of experimental relationships. Section 6.3 describes the new
technique and demonstrates its application to develop a model for the SWCC.
Section 6.4 evaluates the model obtained through its application to a series of
experimental data sets obtained from the literature and for soil obtained from
the field study site (Chapter 4).
The research presented in this chapter has been been reported in a paper
which has been submitted to a peer-reviewed international journal (see List of
publications).
127
128
6.2 Background
The modelling of water and solute transport in porous media and the evalua-
tion of their associated mechanical behaviour has become increasingly signif-
icant as advances in unsaturated soil mechanics continue (Fredlund and Ra-
hardjo, 1993). The soil water characteristic curve (SWCC) is a key relationship
required for this modelling. For a given soil condition, the SWCC represents
the relationship between soil water content and the pore fluid suction at a
given temperature. The SWCC is measured in the laboratory, commonly as
a series of discrete points which for modelling purposes are represented as a
continuous curve by fitting some form of mathematical function. The SWCC
is commonly used to estimate other soil properties, such as shear strength and
unsaturated permeability (Fredlund and Rahardjo, 1993). Accurate descrip-
tion of the SWCC and calculation of its derivative is necessary in order for
reliable prediction of these properties.
The characteristic features of the SWCC are the air-entry value (ψae ), the
residual suction value (ψr ), the minimum suction value (ψmin ), the maximum
suction, the saturated water content (ωsat ), the slope in the initial portion of
the curve (m), the slope in the central portion of the curve (n) and the slope
in the final portion of the curve (o). Figure 6.1 shows an idealised version of
the SWCC with these points identified.
ω Initial portion Central portion Final portion
ω m ϕ
ω
ϕ
ω
o
min
Figure 6.1: Idealised SWCC of volume change soil with labelled features (x-
axis in log scale)
CHAPTER 6. NOVEL MODEL DEVELOPMENT TECHNIQUE 129
Numerous mathematical equations have been used to represent the SWCC

using a general sigmoidal shape (Leong and Rahardjo, 1997, Sillers and Fred-
lund, 2001). Of these, the equations proposed by Brooks and Corey (1964),
van Genuchten (1980) and Fredlund and Xing (1994) are the most commonly
considered (Burger and Shackelford, 2001). Leong and Rahardjo (1997) found
that ’almost all’ equations proposed for the SWCC, including these three can
be derived from a single equation (Equation 6.1) by selection of appropriate
values for the constants.
b b2
a1 Θb1 + a2 ea3 Θ 1 = a4 ψ b2 + aa56 ψ + a7 (6.1)
where a1 , a2 , a3 , a4 , a5 , a6 , a7 , b1 and b2 are constants, ψ is the suction and Θ is

the normalised volumetric water content.
Although these mathematical functions provide continuous curves for math-
ematical modelling, they all suffer from one disadvantage. The majority of
fitting parameters are purely empirical and are not directly related to the char-
acteristic features of the SWCC. The only characteristic points of the SWCC
included in these equations are the intercepts of the curve with the soil water
content and suction axes, being the saturated water content at the minimum
suction and the maximum suction (106 ) at zero water content respectively.
Relationships between different fitting parameters and characteristic points of
the SWCC have been presented; however these are approximate only and are
not applicable in all cases.
In order to create an equation which included fitting parameters directly
related to the critical features of the SWCC, the eight characteristic features
described above, eight fitting parameters are required. In addition to these two
additional fitting parameters are required to account for the curvature of the
SWCC at (ψae ) and (ψr ), φae and φr . The incorporation of fitting parameters
which can be directly related to each feature of the SWCC allows true physical
meaning to be associated with each fitting parameter.
Some of these values can either be readily set manually or can be calculated.
The total number of fitting parameters which are required to be determined
during fitting can be reduced to seven. The maximum suction (defined as 106 )
and ψmin (this value is dependent on experimental results) can be set manually,
and o can be calculated to ensure that the SWCC contacts the suction axis
at the maximum suction. It may be possible to further reduce the number of
fitting parameters by combining or fixing the values of φae and φr , however
this could reduce fitting accuracy.
130
Gitirana and Fredlund (2004) and Pham and Fredlund (2008) both pre-
sented equations which included fitting parameters that were directly related
to the critical features of the SWCC. The equations by Gitirana and Fredlund
(2004) are based on rotated and translated hyperbolas, whilst the equation
by Pham and Fredlund (2008) is based on three individual curves combined
using mathematical functions that transition in value between zero and one as
required. Both equations appear to fit published data well using automated
fitting techniques. However, no metric to describe the goodness of fit was
provided. Where fitting parameters are directly related to critical features of
the SWCC, it should be possible to achieve a good fit using soil properties
determined from the experimental data without relying on automated fitting
techniques. No indication of this is given in either paper. For the remainder
of this chapter this technique will be referred to as manual fitting.
The equations presented by Gitirana and Fredlund (2004) are limited in

application. These equations do not account for the behaviour of high volume
change soil below the air-entry value (Pham and Fredlund, 2008). Additionally,
the equations have been fixed to a suction range between 10−1 and 106 , and
uses a single parameter to control shape of the SWCC at all bending points.
The use of this single parameter assumes that the shape of the SWCC is the
same at these points. Finally, the equations and their first derivatives, whilst
calculable, are both extremely complex.
The equation presented by Pham and Fredlund (2008) does account for
the behaviour of high volume change soil below the air-entry value and allows
for differing rates of change at both bending points. However, this equation
is still highly complex as is its derivative. Pham and Fredlund (2008) did not
provide the derivative of their SWCC equation; however the derivative has
been calculated here and is shown in Section 6.3.5.
In this chapter a single continuous and easily differentiable equation is

developed for the SWCC. The fitting parameters used in the equation are
directly related to the characteristic features of the SWCC. This equation is
developed using a novel method that initially focuses on describing the gradient
of the curve, in contrast to those previously described. The equation presented
is fitted to experimental data using both automated and manual techniques.
This chapter focuses on unimodal SWCCs with two bending points, but the
method presented here can be used to readily develop an equation for multi-
modal curves.
6.3 Equation development technique

The fitting of a curve to a series of experimental data points is a common
practice and is readily achieved where the data follows a simple trend such as
linear or exponential. However, where experimental data exhibits complicated
behaviour this process is more involved. This process becomes more problem-
atic when fitting parameters which are physically relevant are included. The
technique presented here is general and reduces the complexity of the equation
to be fitted, thus simplifying the development process.
This technique has four steps;
1. Plot a curve of the desired shape to fit the data, noting all important
features.
2. Reduce the complexity of the shape of the curve by determining and

plotting the curve’s derivative.
3. Fit an equation to the curve’s derivative.
4. Integrate this equation to arrive at the equation to fit the original data.
To demonstrate the application of this technique, it has been applied to the

SWCC.
6.3.1 Plotting the derivative of the SWCC

It can be seen in Figure 6.1 that the SWCC begins at ψmin (which could be
1 kPa or less to represent zero suction on the log scale) as a line of constant
negative gradient (m). The magnitude of this slope then increases at the air
entry value. Above ψr the magnitude of the slope reduces to a constant value
before the curve intersects the suction axis at 1 × 106 kP a (Fredlund and Xing,
1994). At this point, the water content is considered to be zero. The initial line
of constant negative gradient (m) is an indication of the capillary saturated
compressibility of the soil.
From observation of the SWCC, it is clear that the derivative of the curve
will always be negative. The slope is constant from ψmin , before increasing in
magnitude at ψae . The slope then remains approximately constant until ψr
where the magnitude of the slope then decreases to a final value. The expected
general shape of the derivative of the SWCC is shown in Figure 6.2. The rapid
rate of change at ψae and ψr has been used for clarity only. It should also be
noted that whilst Figure 6.2 shows that the derivative prior to ψae is greater
132
in magnitude than the derivative after ψr this is not a general requirement

for curve development and is only shown here to indicate a difference in the
values.
min
0
m
o
n
dω Initial portion Central portion Final portion
d
Figure 6.2: Idealised derivative of the soil water characteristic curve (x-axis in
log scale)
6.3.2 Equation development

To represent the derivative of the SWCC a function which is able to exhibit
rapid change at two defined points with constant values at other locations
is required. A function which displays two horizontal asymptotes where the
transition between them is definable and has a controllable rate of change at
that point is tan−1 (x). A combination of two tan−1 (x) curves would create
a function which exhibits the desired shape and allow the initial, subsequent
and final values to be set and the location at which the gradient changes to
also be specified.
Following this conceptual understanding, the equation for the derivative of
the SWCC is shown in Equation 6.2.

dω m−n −1 ∗ ∗ n−o −1 ∗ ∗ m+o
∗
=− tan φae (ψ − ψae ) − tan φr (ψ − ψr ) +
dψ π π 2
(6.2)
where
ψ ∗ = log10 (ψ) (6.3)

∗
ψae = log10 (ψae ) (6.4)
ψr∗ = log10 (ψr ) (6.5)
The integration of Equation 6.2 without inclusion of initial conditions re-

sults in Equation 6.6.

∗ m−n
ω(ψ ) = − φae (ψ ∗ − ψae∗
)tan−1 (φae (ψ ∗ − ψae
∗
))
φae .π

1 ∗ ∗ 2
− ln(1 + (φae (ψ − ψae )) )
2
(6.6)
n−o
− φr (ψ ∗ − ψr∗ )tan−1 (φr (ψ ∗ − ψr∗ ))
φr .π

1 ∗ ∗ 2 m+o ∗
− ln(1 + (φr (ψ − ψr )) ) + ψ +C
2 2
The initial condition used to calculate the integration constant is shown in

Equation 6.7.
ω = ωsat when ψ ∗ = ψmin

∗
(ψ = ψmin ) (6.7)
Solving Equation 6.6 for C at this condition results in Equation 6.8.

m−n ∗ ∗ ∗
)tan−1 φae (ψmin ∗

C =ωsat + φae (ψmin − ψae − ψae )
φae .π

1 ∗ ∗ 2

− ln 1 + (φae (ψmin − ψae ))
2
(6.8)
n−o ∗
− ψr∗ )tan−1 φr (ψmin
∗
− ψr∗ )

+ φr (ψmin
φr .π

1 ∗ ∗ 2
m+o ∗
− ln 1 + (φr (ψmin − ψr )) − ψmin
2 2
Combining Equations 6.6 and 6.8 produces the final equation, Equation
6.9.
134

∗ m−n
φae (ψ ∗ − ψae ∗
)tan−1 φae (ψ ∗ − ψae∗

ω(ψ ) =ωsat − )
φae .π

1 ∗ ∗ 2

− ln(1 + (φae (ψ − ψae ))
2

n−o
φr (ψ ∗ − ψr∗ )tan−1 φr (ψ ∗ − ψr∗ )

−
φr .π

1 ∗ ∗ 2

− ln(1 + (φr (ψ − ψr ))
2
(6.9)
m−n ∗ ∗ ∗
)tan−1 φae (ψmin ∗

+ φae (ψmin − ψae − ψae )
φae .π

1 ∗ ∗ 2

− ln(1 + (φ(ψmin − ψae ))
2

n−o ∗
− ψr∗ )tan−1 φr (ψmin
∗
− ψr∗ )

+ φr (ψmin
φr .π

1 ∗ ∗ 2
m+o ∗ m+o ∗
− ln(1 + (φr (ψmin − ψr )) + ψ − ψmin
2 2 2
Equation 6.9 can be rewritten into a more concise form as shown in Equa-
tion 6.10

∗ m−n ∗ ∗ n−o ∗ ∗
ω(ψ ) = − f (ψ ) − f (ψmin ) − g(ψ ) − g(ψmin )
φae .π φr .π (6.10)
m+o ∗ m+o ∗
+ .ψ − .ψmin + ωsat
2 2
where
f (ψ ∗ ) =φae (ψ ∗ − ψae
∗
)tan−1 φae (ψ ∗ − ψae∗

)
1 (6.11)
− ln(1 + (φae (ψ ∗ − ψae ∗
))2 )
2
g(ψ ∗ ) =φr (ψ ∗ − ψr∗ )tan−1 φr (ψ ∗ − ψr∗ )

1 (6.12)
− ln(1 + (φr (ψ ∗ − ψr∗ ))2 )
2
Equation 6.10 is able to be used to represent the SWCC where all fitting
parameters are know. However, the value of o obtained from fitting Equation
6.10 may not ensure that the SWCC contacts the suction axis at the maximum
suction, this requirement is shown in Equation 6.13. To achieve this require-
ment the final slope of the SWCC (o) must be calculated accounting for the
initial behaviour of the curve.
ω = 0 when ψ ∗ = 6 (ψ = 106 ) (6.13)
Solving Equation 6.10 for o at the this condition results in Equation 6.14
m − n ∗
n ∗

f (6) − f (ψmin ) + g(6) − g(ψmin )
φ .π φr .π
o = ae
1
∗
ψ∗
g(6) − g(ψmin ) + 3 − min
φr .π 2
∗ (6.14)
ψmin
−ωsat + − 3 .m
2
+
1 ∗
ψ∗
g(6) − g(ψmin ) + 3 − min
φr .π 2
The value of o can also be approximated using the less complicated Equa-
tion 6.15. However Equation 6.15 does not ensure that the SWCC contacts
the suction axis at 106 exactly.
∗ ∗
ωsat + m(ψae − ψmin ∗
) + n(ψr∗ − ψae )
o∼
=− ∗
(6.15)
6 − ψr
6.3.3 Updated equation
Experimental data for the SWCC, specifically from granular soils, indicates
that the theoretical maximum suction of 106 kPa is not appropriate in all
cases. The SWCC for granular soils has been found to reach a maximum
at much lower values. Based on this observation, the proposed equation has
been updated to allow the maximum suction to be set explicitly or to be
determined during equation fitting. To allow this, Equation 6.14 was updated
to that shown in Equation 6.16. Whilst with Equation 6.16 it is now possible
to vary the value of the maximum suction, for the purposes of the remainder
of this chapter the maximum suction will be set to the theoretical maximum
suction of 106 kPa unless otherwise stated.
136
m − n ∗ ∗
n ∗ ∗

f (ψmax ) − f (ψmin ) + g(ψmax ) − g(ψmin ))
φ .π φr .π
o = ae ψ∗
1
∗ ∗ ψ∗
g(ψmax ) − g(ψmin ) + max − min
φr .π 2 2
∗ ∗
(6.16)
ψmin ψmax
−ωsat + − .m
2 2
+ ψ∗
1 ∗ ∗ ψ∗
g(ψmax ) − g(ψmin ) + max − min
φr .π 2 2
∗
where ψmax = log10 (ψmax ) and ψmax = maximum suction, suction where water
content = 0.
6.3.4 Fitting parameter effect study

The design of the SWCC equations presented here is such that each fitting
parameter is directly related to individual features of the SWCC. This section
demonstrates the behaviour of each fitting parameter. Figures 6.3 to 6.6 show
the effect of changing parameters ψae , ψr , m, n, ωsat , ψmin , φae and φr on
the shape of the SWCC. It can be seen from these figures that each fitting
parameter affects only a single feature of the curve.
Fitting parameters ψae and ψr control the location of the two bending
points in the curve. ψae controls where the curve transitions from the initial
to the central portions of the curve, and ψr controls where the central portion
transitions to the final portion of the SWCC. Where ψae and ψr set to the
same value, the curve would transition directly from the initial to the final
portions of the curve, this has not been shown here.
The role of φae and φr is to dictate the rate of change of the slope around
ψae and ψr respectively. Both φae and φr affect the curve in the same manner.
For this reason, the behaviour of both parameters will be discussed here as
φ. This rate of change is inversely proportional to the value of φ. The result
of this is that as the value of φ decreases the range over which the fitting
parameter influences the shape of the curve increases. When the value of φ is
sufficiently low the rate of change of the SWCC at ψae and ψr is such that the
entire SWCC is affected.
Fitting parameters m and n are the slope of the initial and central portions
of the SWCC respectively. As o is a calculated value, it has not been included
here. However it can be seen that when other fitting parameters are varied,
the value of o is adjusted to ensure the SWCC contacts the suction axis at
1.5
1
ω
0.5 1.5
ψ ae = 10 = 31 kPa
ψ ae = 101 = 10 kPa
ψ ae = 102 = 100 kPa
0
0 2 4 6
10 10 10 10
ψ (kPa)
(a) Varying fitting parameter ψae
1.5
1
ω
0.5 3.5
ψ r = 10 = 3162 kPa
ψ r = 103 = 1000 kPa
ψ r = 104 = 10000 kPa
0
0 2 4 6
10 10 10 10
ψ (kPa)
(b) Varying fitting parameter ψr
Figure 6.3: Parameter sensitivity results (x-axis in log scale). Note: Except
∗
where otherwise stated ψae = 101.5 = 32 kPa (ψae = 1.5), ψr = 103.5 =
3162 kPa, m = −0.025, n = −0.45, ωsat = 1.25, ψmin = 100 = 1 kPa, φae =
5 and φr = 5.
138
1.5
1
ω
0.5
ϕae= 5
ϕae= 2
ϕae= 50
0
0 2 4 6
10 10 10 10
ψ (kPa)
(a) Varying fitting parameter φae
1.5
1
ω
0.5
ϕr = 5
0
0 2 4 6
10 10 10 10
ψ (kPa)
(b) Varying fitting parameter φr
∗
5 and φr = 5.
1.5
1
ω
0.5
m = −0.025
0
0 2 4 6
10 10 10 10
ψ (kPa)
(a) Varying fitting parameter m
1.5
1
ω
0.5
n = −0.45
0
0 2 4 6
10 10 10 10
ψ (kPa)
(b) Varying fitting parameter n
∗
5 and φr = 5.
140
1.5
1
ω
0.5
ωsat = 1.25
ωsat = 1.325
ωsat = 1.4
0
0 2 4 6
10 10 10 10
ψ (kPa)
(a) Varying fitting parameter ωsat
1.5
1
ω
0.5 0
ψ = 10 = 1 kPa
min
ψ -1
= 10 = 0.1 kPa
min
ψ 1
= 10 = 10 kPa
min
0
0 2 4 6
10 10 10 10
ψ (kPa)
(b) Varying fitting parameter ψmin
∗
5 and φr = 5.
106 . It can be seen that changing m does not affect the value of the SWCC
at ψmin . This value was set as ωsat , which is defined above as the saturated
water content and is expected to occur at ψmin . However it can be seen that
if ψmin is not set as the minimum suction shown, these parameters act to set
a fixed point in the curve not located at either axis. While this is possible, it
is not the intended purpose of these fitting parameters.
6.3.5 Equation and derivative from Pham and Fredlund

(2008)
Pham and Fredlund (2008) did not present the derivative of their equation,
shown here in Equation 6.17. The derivative of this equation is given here in
Equation 6.21. It can be seen by comparison of Equations 6.2 and 6.21 that
the derivative of the equation developed here is significantly less complex.

ψ ln(10)
ω(ψ) = A(ψ).(S2 − S1 ) log − [1 − A(ψ)]
ψae 2t1

ψ ln(10)
+ (S3 − S2 ) log − [1 − B(ψ)]) B(ψ) (6.17)
ψr 2t2
6
10
+ S3 .log
ψ
where S1 , S2 and S3 are the initial, centre and final slopes of the SWCC
respectively, t1 and t2 control the transition between slopes,
ωsat + (S2 − S1 )log(ψae ) − S2 log(ψr )

S3 = 6 (6.18)
10
log
ψr
t1
ψ
A(ψ) = t ae t1 (6.19)
ψ 1 + ψae
ψrt2
B(ψ) = − t (6.20)
ψ 2 + ψrt2
142

dω (S2 − S1 ) A(ψ).B(ψ) 0 ψ
= + A (ψ).B(ψ).ln
dψ ln(10) ψ ψae

ψ
+ A(ψ).B 0 (ψ).ln

ψ
ae
ln(10)(S2 − S1 ) 0
− A (ψ).B(ψ) + A(ψ).B 0 (ψ)
2t1
(6.21)
0 0
− A2 (ψ).B(ψ) − A2(ψ).B (ψ)

(S3 − S2 ) 0 ψ B(ψ)
+ B (ψ).ln +
ln(10) ψr ψ

(S3 − S2 ).ln(10) 0 0 S3
+ B2 (ψ) − B (ψ) −
2t2 ln(10).ψ
where
ψ (t1 −1) .ψae

t1
.t1
A0 (ψ) = − t (6.22)
(ψ t1 + ψae )21
ψ (t2 −1) .ψrt2 .t2

B 0 (ψ) = − t (6.23)
(ψ 2 + ψrt2 )2
2t1
ψae
A2(ψ) = A(ψ).A(ψ) = t1
2 (6.24)
ψ t1 + ψae
2ψ (t1 −1) .ψae
2t1
.t1
A20 (ψ) = − t (6.25)
(ψ t1 + ψae1 )3
ψr2t2
B2(ψ) = B(ψ).B(ψ) = 2 (6.26)
ψ t2 + ψrt2
2ψ (t2 −1) .ψr2t2 .t2
B20 (ψ) = − (6.27)
(ψ t2 + ψrt2 )3
6.4 Evaluation of SWCC equation using ex-

perimental data
SWCC data were compiled from the literature that represent a wide range
of soil types for evaluation of the proposed equation. SWCC data were also
obtained using soil obtained from the Altona North field study site (Chapter 4)
by Chan (InPrep). Chan (InPrep) obtained SWCC data for soil at two depths;
from ground level (0 mm) to 350 mm depth and 350 mm to 500 mm depth. The
SWCC data were fitted to the proposed equation using both automated and
manual fitting techniques. Automated fitting and evaluation was undertaken
using the Gauss-Marquardt-Levenberg algorithm incorporated into the PEST

software program (Doherty, 2003). Manual fitting was undertaken by using
values determined directly from the raw data. φa and φr were initially set
as 15, these values were then adjusted as required to achieve a visually close
fit. As some of this data was presented using the degree of saturation (Sr ) or
volumetric water content (Θ), Sr or Θ replaces ω as appropriate.
Both automated and manual fitting results are shown in Figures 6.7 to
6.14. The goodness of fit was determined using the coefficient of determination
(R2 ) as the metric. Table 6.1 shows a comparison of R2 values for manual
and automated fittings. It can be seen in Table 6.1 that the goodness of
fit achieved using manually determined fitting parameters is only marginally
lower than that achieved by automated fitting. The results achieved by manual
fitting highlights the direct relationship between fitting parameters and critical
features of the SWCC.
The raw data used in equation fitting, the results of automated fitting and
the results of manual fitting are shown as circles, solid lines and dashed lines
respectively. Several data points were excluded from the fitting of soil from the
top 350 mm due to non-conformity and these points are shown as red circles
(Figure 6.13).
Table 6.1: Automated vs. Manual fitting results

Soil R2 of Automated R2 of Manual
Fitting Fitting
Silty Loam 0.999 0.987
Silty Clay 1.0 0.999
Sandy Loam 0.998 0.994
Kidd Creek Tailings 0.997 0.997
Jossigny Loam 0.997 0.996
Initially Slurried Regina Clay 0.997 0.995
Altona North - 0 - 350 mm 0.985 0.966
Altona North - 350 - 500 mm 0.988 0.987
144
Raw data Automated Manual

ψa = 34 kPa
0.25
ψr = 158 kPa
m = −0.001 m = −0.004
n = −1.33 n = −0.20
ωsat = 0.27 ωsat = 0.27
ɸae = 4.45 ɸae = 5
0.2
ω − Gravimetric Water Content
ɸr = 5.25 ɸr = 5
ψmin =0.01 kPa ψmin =0.01 kPa
2
R = 0.999 R2 = 0.987
0.15
0.1
0.05
0
−2 −1 0 1 2 3 4 5 6
10 10 10 10 10 10 10 10 10
Figure 6.7: Silty Loam (Raw data from University of Saskatchewan, personnel
communication S.L. Barbour (2004) (x-axis in log scale)
0.4
ψa = 24 kPa
ψr = 1×105 kPa
0.35
m = −0.008 m = −0.01
n = −0.09 n = −0.08
ωsat = 0.38 ωsat = 0.38
0.3 ɸae = 4.6 ɸae = 10
ɸr = 1000 ɸr = 15
ψmin =0.1 kPa ψmin =0.1 kPa
0.25 R2 = 1.0 R2 = 0.999
0.2
0.15
0.1
0.05
0
−1 0 1 2 3 4 5 6
10 10 10 10 10 10 10 10
Figure 6.8: Silty Clay (Raw data from University of Saskatchewan, personnel
communication S.L. Barbour (2004) (x-axis in log scale)
146
0.5
ψae = 1.3 kPa ψae = 1.2 kPa
0.45 ψr = 6.1 kPa ψr = 6.6 kPa
m = 0.0 m = 0.0
n = −0.51 n = −0.451
0.4 ωsat = 0.5
ωsat = 0.50
ɸae = 4.53 ɸae = 7
0.35 ɸr = 4.76 ɸr = 8
ψmin = 0.01 kPa ψmin = 0.01 kPa
R2 = 0.998 R2 = 0.994
0.3
0.25
0.2
0.15
0.1
0.05
0
−2 −1 0 1 2 3 4 5 6
10 10 10 10 10 10 10 10 10
Figure 6.9: Sandy Loam (Raw data from University of Saskatchewan, person-
nel communication S.L. Barbour (2004) (x-axis in log scale)
1
Sr − Degree of Saturation
0.8
0.6
Automated Manual Raw data

ψa = 993 kPa ψ a = 4154 kPa
0.4
ψr = 8407 kPa ψr = 1×106 kPa
m = 0.0 m = −0.006
n = −0.17 n = −0.41
Srsat= 1.0 Srsat= 1
0.2 ɸae = 16.3 ɸae = 15
ɸr = 1000 ɸr = 15
ψmin = 1 kPa ψmin = 1 kPa
2
R = 0.997 R2 = 0.995
0
0 1 2 3 4 5 6
10 10 10 10 10 10 10
Figure 6.10: Regina Clay (Raw data from Fredlund and Xing (1994)) (x-axis
in log scale)
148

ψa = 241 kPa ψa = 128 kPa
ψr = 477 kPa ψr = 893 kPa
m = 0.0 m = −0.002
1 n = −2.94 n = −0.109
Srsat= 1.0 S rsat= 1.0
ɸae = 6.02 ɸae = 5.5
ɸr = 6.92 ɸr = 5
Sr − Degree of Saturation
0.8 ψmin = 0.1 kPa ψmin = 0.1 kPa

2
R = 0.999 R2 = 0.997
0.6
0.4
0.2
0
−1 0 1 2 3 4 5 6
10 10 10 10 10 10 10 10
Figure 6.11: Kidd Creek Tailings (Raw data from Fredlund and Xing (1994))
(x-axis in log scale)
0.5
ψae = 9.88 kPa ψae = 2 kPa
0.45 ψr = 6224 kPa ψr = 10249 kPa
m = −0.08 m = 0.0
n = −0.12 n = −0.10
0.4 ωsat = 0.54 ωsat = 0.46
ɸ ae = 1000 ɸ ae = 50
0.35 ɸr = 1 ɸ r = 50
2
R = 0.999 R2 = 0.996
0.3
0.25
0.2
0.15
0.1
0.05
0
−1 0 1 2 3 4 5 6
10 10 10 10 10 10 10 10
Figure 6.12: Jossigny Loam (Raw data from Fleureau et al. (2002)) (x-axis in
log scale)
150
0.5
ψae = 0.29 kPa ψae = 0.44 kPa
0.45 ψr = 2573 kPa ψr = 1585 kPa
m = 0.0 m = 0.0
n = −0.09 n = −0.1
0.4
Θsat = 0.48 Θsat = 0.48
ɸae = 1000 ɸae = 50

θ − Volumetric Water Content
0.35 ɸr = 882 ɸr = 50
R2 = 0.985 R2 = 0.966
0.3
0.25
0.2
0.15
0.1
0.05
0
−2 −1 0 1 2 3 4 5 6
10 10 10 10 10 10 10 10 10
Figure 6.13: Field site soil from ground level to 350 mm depth (Raw data from
Chan (InPrep)) (x-axis in log scale)
0.5
0.45
0.4
θ − Volumetric Water Content
0.35
0.3
0.25
0.2 Raw data Automated Manual

ψae = 345 kPa ψae = 322 kPa
ψr = 22232 kPa ψr = 39602 kPa
0.15
m = −9.59E−03 m = −0.01
n = −1.52E−01 n = −0.14
Θsat = 0.45 Θsat = 0.45
0.1
ɸae = 114 ɸae = 50
ɸr = 77.7 ɸr = 50
0.05 ψmin = 0.01 kPa
ψmin = 0.01 kPa
R2 = 0.988 R2 = 0.987
0
−2 −1 0 1 2 3 4 5 6
10 10 10 10 10 10 10 10 10
Figure 6.14: Field site soil from 350 mm to 500 mm depth (Raw data from
Chan (InPrep)) (x-axis in log scale)
152
6.5 Discussion
Figures 6.7 to 6.14 show that the proposed equation is able to fit experimental
data over a wide range of soil types. It can also be seen that the equation
is not limited in application to either gravimetric water content or degree of
saturation, but is equally applicable to both variables.
The development of the proposed equation by first describing the gradient
of the SWCC has ensured that the derivative of the equation can be explicitly
determined, an important feature for numerical modelling. Additionally the
equation chosen to describe the derivative of the SWCC is less complex than
the SWCC equation itself and is also much less complex than the derivatives
of the equations proposed by (Gitirana and Fredlund, 2004) and Pham and
Fredlund (2008).
Although the equation proposed for the SWCC also looks complex, it is
based on a mathematical pattern and once the technique used in equation
development is understood, it becomes possible to develop similar equations
for even more complex behaviours such as multimodal SWCCs.
Finally when applying automated fitting algorithms to equations such as
the one presented here, the results produced should always be manually checked
as non-uniqueness of the fitting parameters is possible. The output of auto-
mated fitting algorithms are fitting parameters which best fit the input data.
However these fitting parameters may not be truly representative of real be-
haviour. An example of this can be observed in relation to the fitting of the
proposed equation to the experimental data for Jossigny Loam (Figure 6.12).
Automated fitting produced an air-entry value of 9.9 kPa. However Fleureau
et al. (2002) indicates that the true air-entry value is around 740 kPa. Refit-
ting of the equation using this value for the air-entry value resulted in only a
minor drop in the goodness of fit since R2 only reduced from 0.999 to 0.998.
Due to the shape of SWCC, varying the air-entry value between 10 and 1000
resulted in a reduction of the goodness of fit, measured as R2 , of only 0.001.
6.6 Conclusions
This chapter has presented a novel technique for the creation of empirical mod-
els. Models are developed by first determining an equation for the derivative
of the final model equation. The use of the derivative as the starting point
simplifies the inclusion of fitting parameters which are directly related to the
characteristic features of the relationship modelled. This also ensures that the
derivative of the model is simple to obtain.

In this chapter this technique was applied to the unimodal form of the
SWCC and is shown to produce a flexible and reliable model. However this
method can be readily applied to produce an equation to describe soils with
multimodal SWCCs. The model created achieved an excellent fit to all datasets
used in the validation. The determination of fitting parameters was conducted
using both automated fitting software and directly from the experimental data.
Both techniques produced excellent fitting results with R2 values in excess of
0.98 in all cases. It has been shown that fitting parameters determined directly
from the experimental data produce excellent results, which are extremely
close to those achievable by automated fitting, whilst allowing sensible physical
parameters to be explicitly imposed.
Chapter 7
A void ratio - water content -

net stress model for
environmentally stabilised
expansive soils
7.1 Introduction
Chapter 6 presented a technique for the development of empirical model equa-

tions. This chapter applies this technique to the relationship between soil void
ratio and soil water content of environmentally stabilised soils. The model
is then extended to incorporate the effect of net stress. These models will
provide a practical approach for the prediction of soil strain/displacement, in
expansive soils, resulting from changes in water content at a range of stress
conditions, particularly for use in 1-D or isotropic analysis.
Section 7.2 provides background on the relationship between soil void ratio
and soil water content and net stress. Section 7.3 presents a new mathematical
model for the soil shrinkage curve. The equation is extended in Section 7.4
to incorporate the effect of net stress. Section 7.5 discusses the theoretical
associations of the empirical fitting parameters with soil mechanics theory.
The work presented in this chapter has been reported in a paper published
in a peer-reviewed international journal (see List of publications).
155
156
7.2 Background
Volume change in expansive soils is a widely known behaviour resulting from
changes in soil water content. This volume change has a significant affect on
surface and buried infrastructure with high economic and social consequences.
It was reported by Gould et al. (2009) that the shrink/swell behaviour of soils
has a significant affect on the failure rates of buried water reticulation pipes.
Vu (2002) reported that annually expansive soils in the United States cause
more damage to structures than natural disasters.
The degree to which volume change occurs in soil is strongly dependant on
several factors, soil mineralogy (primarily clay content and type), the change
in soil water content, suction, initial conditions and net stress. Soil miner-
alogy, particularly the clay percentage and type, influences volume change
of soil as this behaviour results from the shrink/swell behaviour of the clay
fraction within the soil. The higher the percentage of clay and the greater
the shrink/swell potential of that clay the higher the volume change observed
(Thomas et al., 2000). Although the actual relationship is more complex than
described here, this is generally applicable.
Initial conditions affect volume change in two primary ways. The first
relates to the initial water content. If the water content is below the shrinkage
limit no appreciable swelling is observed until the gravimetric water content
exceeds this value (referred to as water content herein). The reverse is also true
when the initial water content is above the swelling limit, i.e. no appreciable
shrinkage is observed until the water content reduces to below this value. The
second relates to the history of the soil. Specimens subjected to conditions
outside of previous experience, either net stress and/or water content have
been shown to exhibit irreversible plastic behaviour due to these conditions
(Wheeler et al., 2003). However soil that has undergone a series of wetting
and drying cycles, becoming ripened soil (Kodikara et al., 2002), appears to
show reversible behaviour (Tripathy et al., 2002).
The change in volume exhibited by a soil due to water content change
is suppressed by net stress. As net stress conditions increase, the possible
volume change exhibited by a soil reduces. Extensive testing was undertaken
by Tripathy et al. (2002) in which remoulded soil samples were subject to
consecutive wetting and drying cycles under three different levels of net stress.
This work clearly demonstrated the changes imposed on the shrinkage curve
due to the influence of net stress conditions. A similar observation was also
reported by Groenevelt and Bolt (1972).
CHAPTER 7. VOID RATIO-WATER CONTENT-NET STRESS 157
The prediction of soil volume change requires the use of a governing pa-
rameter which drives volume change, of which there are two possible choices,
namely soil water content or soil suction (Briaud et al., 2003). These govern-
ing parameters are related via the soil water characteristic curve (SWCC) and
therefore for a single soil type the use of either parameter is possible as they
are readily translated into each other, provided net stress is constant. This is
supported by Groenevelt and Bolt (1972) who stated that a swelling soil sys-
tem is characterised by void ratio, moisture ratio (water content), suction and
net stress, of which two parameters can be considered independent. Therefore
for any chosen value of net stress there exists a single void ratio - water con-
tent relationship that fully determines the suction. The consequence of this is
that it is possible to characterise a swelling soil using net stress, water content
and void ratio without the need to determine suction. Alternatively all four
parameters can be incorporated through the use of two constitutive surfaces
(Vu and Fredlund, 2004).
The use of suction as a governing parameter introduces the presence of

hydraulic hysteresis between wetting and drying cycles (Chu and Mou, 1973).
Hydraulic hysteresis is unavoidable when suction is used as a governing param-
eter. This is due to the difference in the critical radii during the flooding or
draining of a void during wetting or drying respectively. The critical radius for
draining a void is smaller than the critical radius for flooding a void, resulting
in the suction for any degree of saturation being higher during drying than
during wetting (Buisson and Wheeler, 2000). Unlike the relationship between
suction and void ratio, the water content - void ratio relationship does not
exhibit hydraulic hysteresis. This was clearly demonstrated with experimental
data by Fleureau et al. (2002), who reported results from simultaneous mea-
surements of void ratio, water content and suction. As can be seen in Fig.
7.1, clear hysteresis exists between wetting and drying pathways for suction,
but no such hysteresis can be seen for water content. Tripathy et al. (2002)
also found no indication of hysteresis in remoulded soil samples after they had
been subjected to four cycles of wetting and drying. Some hysteresis was seen
in initial cycles before the soil had ripened.
The use of soil water content over suction also allows the simplification of
experimental procedures. Soil water content can be measured continuously
and reliably outside of the range for which continuous suction measurements
are possible, also the response of soil water content sensors is much faster and
less problematic than soil suction sensors (Tarantino et al., 2008).
158
r = 100%
Drying
g
in
ry
D
&
g
tin
et
W
Wetting
Suction (kPa)
Figure 7.1: Comparison of Water Content and Suction Curves. Adapted from
Fleureau et al. (2002)
The use of the soil water content as the governing parameter for soil vol-
ume change, whilst not allowing for the explicit inclusion of independent stress
states, is a practical method which effectively sidesteps the complications aris-
ing from suction measurement and hydraulic hysteresis.
The volume change behaviour of soil as water content reduces has been
studied extensively in the field of soil science with the resulting curve known
as the shrinkage curve, an idealised version is shown in Fig. 7.2. The shrinkage
curve has been segregated into zones, however the number of zones is not
clearly defined as some sources have described four zones (Reeve and Hall,
1978, Braudeau et al., 1999), whilst others describe only three (Tripathy et al.,
2002). The three common zones are the structural shrinkage, proportional or
normal shrinkage and residual shrinkage zones were identified. Where four
zones are described an additional zero shrinkage zone is included. For the
purposes of this chapter the shrinkage curve will be defined as containing three
possible zones.
These zones are normally described in relation to soil shrinkage (Braudeau

et al., 1999, Groenevelt and Grant, 2004, Olsen and Haugen, 1998, Peng and
Horn, 2005, Reeve and Hall, 1978, Talsma, 1977, Tripathy et al., 2002). How-
ever, as discussed above, the volume change - water content relationship does
not seem to exhibit hysteresis, therefore these zones are observable during
both shrinkage and swelling behaviour. For this reason these zones have been
renamed as the structural change, proportional change and residual change
zones.
2.0
Structural
Change
ɸ
1.5
Proportional m
Change
1.0 .G S
e
Residual ω
ɸ =
Change ,e
00%
1
=
S
r
0.5
er
b a
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

ω
Figure 7.2: Idealised shrinkage curve
7.2.1 Previous work

Previous work to mathematically describe the shrinkage curve has produced
several equations. A summary of published equations is given in Table 7.1.
It can be seen that, with the exception of the equation proposed by Peng
and Horn (2005), no single equation is able to describe all three zones of the
shrinkage curve. Multiple equations, acting over discrete ranges, are combined
to describe the shrinkage curve. The equation proposed by Peng and Horn
(2005) has been applied to data reported in the literature with some success.
However, whilst showing good fitting to laboratory data, the design of this
equation is such that the fitting parameters cannot be related to specific fea-
tures of the shrinkage curve.
160
Table 7.1: Comparison of shrinkage curve equations

Zones described
Equation Number of equations
SC† PC§ RC‡
Groenevelt and Bolt (1972) × X X 1
Olsen and Haugen (1998) X X X 2
Braudeau et al. (1999) X X X 5
Chertkov (2000) × X X 3
Fredlund et al. (2002) × X X 1
Peng and Horn (2005) X X X 1
† § ‡
Structural Change, Proportional Change, Residual Change
7.3 Shrinkage curve equation definition
Equation 7.1 calculates void ratio from gravimetric water content. This equa-
tion is referred to as the void ratio - water content (e − ω) equation for the
remainder of this chapter. As the e − ω equation was developed using the
technique presented in Chapter 6. This technique has ensured that the fit-
ting parameters are directly related to each feature of the shrinkage curve.
The direct relationship between each fitting parameter and each feature of the
shrinkage curve is shown in Section 7.3.1, by conducting a parameter affect
study.
The equation is shown in two parts for the purpose of readability only.
As noted earlier, the gravimetric water content has been used to decouple the
volumetric affects from water content.
e(ω) = f (ω, a) − f (ω, b) − f (0, a) + f (0, b) + er (7.1)
where

m −1 1 2
f (x, y) = − φ(x − y)tan (φ(x − y)) − ln(1 + (φ(x − y)) ) (7.2)
φ.π 2
e is the void ratio, w is the gravimetric water content, m, a, b, er and φ are

fitting parameters. x is replaced with either ω or 0 as appropriate. y is replaced
with either a or b as appropriate.
Equation 7.1 is limited to the range of gravimetric water contents which
satisfy the inequality in Equation 7.3. This inequality excludes any water
contents which would result in the void ratio being incorrectly calculated above
100% saturation.
e ≥ Gs .ω (7.3)
where Gs is specific gravity.

The derivative of the refined equation is shown in Equation 7.4.
de m m
= − tan−1 (φ(ω − a)) + tan−1 (φ(ω − b)) (7.4)
dω π π
Detail on the steps taken in the derivation of the refined form of Equation
7.1 is given in Appendix C.
7.3.1 Parameter affect study

The design of the e − ω equation presented here is such that each fitting
parameter is directly related to specific features of the shrinkage curve. This
section demonstrates the effect of each fitting parameter. Fig. 7.3 shows the
effect of changing parameters a, b, er , φ and m on the shape of the shrinkage
curve. The features to which each fitting parameter is associated are shown
in Fig. 7.2. It can be seen that each parameter affects only a single feature of
the curve.
Fitting parameters a and b control the water contents at which the magni-
tude of the e − ω curve’s slope decreases and increases respectively. The role of
φ is to dictate the rate of change of the slope around points a and b. This rate
of change is inversely proportional to the value of φ. The result of this is that
as the value of φ decreases the range over which the parameter influences the
shape of the curve increases. When the value of φ is sufficiently low the rate of
change of the shrinkage curve around a and b is such that the entire shrinkage
curve is affected by this feature. The er fitting parameter can be seen to shift
the curve vertically. Finally m affects the gradient of the sloped section (pro-
portional change zone) of the shrinkage curve. The theoretical considerations
associated with each fitting parameter are discussed in Section 7.5.1.
The direct relationship between each fitting parameter and the features
of the shrinkage curve seen here has not been realised in the other proposed
equations. Parameter sensitivity results reported by Peng and Horn (2005)
revealed that only two of the five parameters in their equation could be directly
related to features of the curve, being the residual void ratio and saturation
void ratio. The remaining fitting parameters affected the shape of the entire
curve and could not be assigned to any specific features. An analysis conducted
on the equation proposed by Groenevelt and Bolt (1972) revealed a similar
result where all parameters, with the exception of the saturation void ratio,
affected the entire curve and could not be assigned to any specific features.
(a) (b) (c)
3 3 3
2.5 2.5 2.5
2 2 2
1.5 1.5 1.5
e
e
1 1 1
0.5 0.5 0.5
0 0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
ω ω ω
(d) (e)
3 3
2.5 2.5
2 2
1.5 1.5
e
e
1 Sr = 100% 1 Sr = 100%
ɸ = 200 m = 2.7
0.5 ɸ = 35 0.5 m = 2.1
ɸ = 15 m = 1.5
0 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1
ω ω
Figure 7.3: Parameter Sensitivity. Note: Except where otherwise stated a = 0.8, b = 0.2, er = 1.0, φ = 200 and m = 2.7. (a) the
162
effect of changing fitting parameter a, (b) the effect of changing fitting parameter b, (c) the effect of changing fitting parameter er ,
(d) the effect of changing fitting parameter φ and (e) the effect of changing fitting parameter m
7.3.2 Evaluation e − ω equation using published data

The data published by Olsen and Haugen (1998), Peng and Horn (2005), Reeve
and Hall (1978), Talsma (1977) and Tripathy et al. (2002) were fitted to the
proposed e − ω equation using the PEST software program (Doherty, 2003).
A summary of the properties of the soil used by these authors is given in Table
7.2. More details on these soils can be found in the respective papers. The
original data and fitted equations are shown as points and lines respectively.
The legend in each figure gives the values of the fitting parameters, Figs. 7.4,
7.5, 7.6, 7.7, 7.8, 7.9 and 7.10.
R2 values of the fitted equation for each data set are shown in Table 7.3.
Where the initial fitting of the equation to a data set resulted in a negative
value for b, this value was set to zero and the equation refitted. This was
the only restriction imposed on the value of any fitting parameters. Negative
values for b were found to occur when no residual change zone was observable
in the data.
7.3.3 Comparison to other equations

A comparison for the fitting results for the proposed e − ω equation and the
equations published by Peng and Horn (2005), Olsen and Haugen (1998) and
Groenevelt and Bolt (1972) is shown in Table 7.3. It can be seen that all
four equations fit the data with R2 values exceeding 0.9 for all equations. The
e − ω equation proposed here fit all datasets, achieving an R2 value in excess
of 0.989 in all cases and exceeding 0.995 in the majority of cases (> 80%).
The separation of φ into two parameters, φa and φb associated with a and b
respectively, can improve fitting. This was not implemented here as the small
improvement possible does not justify the introduction of an additional fitting
parameter.
7.3.4 Comparison of automated fitted results to manual

fitting of the e − ω equation
The major improvement of the proposed e − ω equation over other equations
is that each fitting parameter is directly related to specific features of the
shrinkage curve. This feature allows manual fitting of the equation to be
undertaken with a minimum of effort. In order to demonstrate this the e − ω
equation was manually fitted to two sets of data exhibiting three zones of
soil behaviour. The values for m, a, b and er were measured directly from the
Table 7.2: Soil Properties
Plastic Shrinkage Specific Dry Density Clay Content Silt Sand Description/
Data
Limit Limit Gravity (g.cm−3 ) (< 2µm) Content Content Classification
Olsen and Haugen
29% 26% 2.60 - 36% 40% 24% Typic Endoaqualf
(1998) - 0-20 cm
Olsen and Haugen
27% 26% 2.71 - 49% 36% 15% Typic Endoaqualf
(1998) - 20-40 cm
Peng and Horn (2005)
- - - - - - - Typic Chromexert
- Plots A,B & C
Reeve and Hall Strong fine angular
- - - 1.08 58% 35% -
(1978) - Wyre Bw blocky
Moderate medium
Reeve and Hall
- - - 1.13 64% 31% - prismatic (breaking
(1978) - Wyre Bg
to angular blocky)
Reeve and Hall Moderate coarse
- - - 1.03 59% 26% -
(1978) - Fladbury Bg angular blocky
Reeve and Hall (1978) Moderate fine angular
- - - 1.47 36% 33% -
- Faulkbourne Bw blocky
Moderate medium
Reeve and Hall (1978)
- - - 1.46 44% 33% - prismatic (breaking
- Faulkbourne Btg
to angular blocky)
Reeve and Hall Moderate coarse
- - - 1.48 44% 24% -
(1978) - Ragdale Bg angular blocky
Talsma (1977) - All - - 2.69 - 60% - - Black Earth
Tripathy et al. (2002) Processed expansive
42% 10.6% 2.68 - 62% 36% 2%
- Soil A soil
164
Tripathy et al. (2002) Processed expansive

32% 13.5% 2.73 - 52% 28% 20%
- Soil B soil
Table 7.3: Comparison of equation using R2
Olsen and Peng and Horn Groenevelt and
Data Set e − ω Equation §
Haugen (1998) (2005)† Bolt (1972)‡
Olsen and Haugen (1998) - 0-20 cm 0.992 0.982 - 0.969
Olsen and Haugen (1998) - 20-40 cm 0.997 0.995 - 0.991
Peng and Horn (2005) - Plot A 0.997 - 0.996 -
Peng and Horn (2005) - Plot B 0.990 - 0.990 -
Peng and Horn (2005) - Plot C 0.996 - 0.998 -
Reeve and Hall (1978) - Wyre Bw 0.999 - 0.996 0.985
Reeve and Hall (1978) - Wyre Bg 0.999 - 0.998 0.995
Reeve and Hall (1978) - Fladbury Bg 0.999 - 1.000 0.999
Reeve and Hall (1978) - Faulkbourne Bw 0.997 - 1.000 0.989
Reeve and Hall (1978) - Faulkbourne Btg 0.992 - 0.998 0.999
Reeve and Hall (1978) - Ragdale Bg 0.999 - 0.998 0.999
Talsma (1977) - 0.2 mBar 0.995 - 0.992 0.911
Talsma (1977) - 1.4 mBar 0.989 - - -
Talsma (1977) - 63 mBar 0.998 - 0.998 0.998
Talsma (1977) - 112 mBar 0.999 - 1.000 0.978
Tripathy et al. (2002) - Soil A 6.25 kPa 0.996 - - -
Tripathy et al. (2002) - Soil A 50 kPa 0.999 - - -
Tripathy et al. (2002) - Soil A 100 kPa 0.998 - - -
Tripathy et al. (2002) - Soil B 6.25 kPa 0.998 - - -
Tripathy et al. (2002) - Soil B 50 kPa 0.996 - - -
CHAPTER 7. VOID RATIO-WATER CONTENT-NET STRESS
† Fitting results from Peng and Horn (2005). R2 calculated from reported correlation coefficient (r)
§ Fitting results published in Olsen and Haugen (1998)
‡ R2 calculated using fitting parameters published in Groenevelt and Grant (2001), Groenevelt et al. (2001) and Groenevelt and
Grant (2002) and data obtained from original papers, Talsma (1977), Reeve and Hall (1978) and Olsen and Haugen (1998).
165
166
0−20 cm: a = 3, b = 0.29, e0 = 0.8, ɸ = 25.5 & m = 4.37 − R2= 0.996
Figure 7.4: e − ω equation fit to Olsen and Haugen (1998)
published data using a ruler, φ was initially set as 250. The value of φ was
then adjusted as required to achieve a visually close fit. Table 7.4 shows a
comparison of fitting parameters determined manually and those determined
by automated fitting as described above. It can be seen that no significant
difference in parameters exist and R2 values for the manually fitted curves
are comparable to those achieved by automated fitting. The shrinkage curves
produced by the values shown in Table 7.4 are shown in Fig. 7.11 alongside
results from automated fitting.
1.4
1.2
0.8
e
0.6
0.4
Plot A: a = 0.27, b = 0.16, e0 = 0.53, ɸ = 13.1 & m = 8.19 − R2= 0.998

0.2
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45
ω
Figure 7.5: e − ω equation fit to Peng and Horn (2005)
Table 7.4: Automated vs. Manual fitting results

Fitting Automated Manual
Data
Parameter Result Value
a 0.46 0.46
b 0.08 0.09
Tripathy et al. (2002) er 0.58 0.6
Soil A 6.25 kPa φ 33.3 200
m 3.13 2.86
R2 0.998 0.995
a 0.30 0.30
b 0.11 0.10
Tripathy et al. (2002) er 0.48 0.48
Soil A 100 kPa φ 100 250
m 2.65 2.42
2
R 0.998 0.998
168
1.6
1.4
1.2
0.8
e
0.6
0.4
Wyre Bw: a = 0.49, b = 0.14, e0 = 0.63, ɸ = 143.4 & m = 2.39 − R2= 0.999
0.2
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
ω
Figure 7.6: e − ω equation fit to Reeve and Hall (1978) (soils 1 to 3 of 6)

0.9
0.8
0.7
0.6
0.5
e
0.4
0.3
0.2
Faulkbourne Bw: a = 0.21, b = 0.13, e0 = 0.41, ɸ = 16.9 & m = 5.67 − R2= 0.997
0.1
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35
ω
Figure 7.7: e − ω equation fit to Reeve and Hall (1978) (soils 4 to 6 of 6)

170
1.8
1.6
1.4
1.2
1
e
0.8
0.6
0.2 mBar: a = 1.9, b = 0.11, e0 = 1.3, ɸ = 1000 & m = 1.03 − R2= 0.996
0.4
0.2
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
ω
Figure 7.8: e − ω equation fit to Talsma (1977)

6.25 kPa: a = 0.46, b = 0.08, er = 0.58, ɸ = 33.2 & m = 3.13 − R2= 0.996
50 kPa: a = 0.35, b = 0.10, er = 0.52, ɸ = 419 & m = 2.64 − R2= 0.999
100 kPa: a = 0.30, b = 0.11, er = 0.48, ɸ = 100 & m = 2.65 − R2= 0.998
Figure 7.9: e − ω equation fit to Tripathy et al. (2002) soil A

172
1.4
1.2
0.8
e
0.6
0.4
0.2 6.25 kPa: a = 0.38, b = 0.1, e0 = 0.57, ɸ = 352.2 & m = 2.49 − R2= 0.999
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
ω
Figure 7.10: e − ω equation fit to Tripathy et al. (2002) soil B

1.8
1.6
1.4
1.2
1
e
0.8
0.6
0.4 6.25 kPa Auto: a = 0.46, b = 0.08, e0 = 0.58, ɸ = 33.3 & m = 3.13 − R2= 0.998
0.2
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
ω
Figure 7.11: Automated vs. Manual fitting results

174
7.4 Shrinkage surface equation definition
The equation proposed in Equation 7.1 is extended to incorporate net stress

such that void ratio is calculated from gravimetric water content and net stress.
The data published by Tripathy et al. (2002) exhibited all three zones of soil
behaviour and were obtained at three net stress levels. For this reason these
data were applicable for fitting to the e − ω − σ equation. The e − ω equation
was initially fitted to this data, as shown in Fig. 7.9. Analysis of the fitting
parameters found significant variation for fitting parameters a and er between
data sets. No significant variation was seen for fitting parameters m, b and φ.
The values found for a and er were then fitted with a logarithmic trendline
of the form commonly used to represent the response of soil to net stress
with good correlation. On the basis of this analysis the e − ω equation was
extended to incorporate net stress via Equations 7.5 and 7.6 which replace
fitting parameters a and er respectively.

σ
a = αa ln 1 + + a0 (7.5)
σ0

σ
er = αer ln 1 + + er0 (7.6)
σ0
where σ0 is the nominal net stress, and αa , a0 , αer and er0 are fitting
parameters.
The partial derivatives of the refined e − ω − σ equation with respect to ω

and σ are shown in Equations 7.7 and 7.8 respectively.

∂e m −1 σ m
= − tan (φ(ω − (αa ln 1 + + a0 ))) + tan−1 (φ(ω − b)) (7.7)
∂ω π σ0 π

∂e αa .m −1 σ
=− .tan −φ αa ln 1 + + a0
∂σ π.σ0 1 + σ0σ σ0

αa .m −1 σ
+ .tan φ ω − αa ln 1 + + a0 (7.8)
π.σ0 1 + σσ0 σ0
αer
+
σ0 1 + σσ0
7.4.1 Evaluation of e − ω − σ equation using published

data
The data from Tripathy et al. (2002) were fitted to the proposed e−ω−σ equa-
tion using the PEST software program (Doherty, 2003). The original data are
shown as circle points, whilst the lines of constant e, ω and σ calculated from
the equation are shown as solid lines (Fig. 7.12). As can be seen the proposed
equation fit the experimental data, achieving an R2 value of 0.997. Figs. 7.13,
7.14 and 7.15 present 2D projections of this surface showing lines of constant
void ratio, stress and water content respectively. The partial derivatives with
respect to ω and σ are shown in Figure 7.16 for a range of σ and ω values
respectively.
7.4.2 Use of compression index in the constant e − ω − σ

equation
In situations where compression index (CC ) of a soil sample is known, it is
possible to replace the a fitting parameter, as calculated in Equation 7.5, with
an approximation using CC . Using the form of the CC equation as shown in
Equation 7.9, a can be approximated as shown in Equation 7.10.

σ
esat = e0 + CC log10 (7.9)
σ0
esat − er
a≈ +b (7.10)
m
where es at is the saturated void ratio, e0 is the nominal saturated void ratio
and CC is the compression index.
Whilst this use of CC is an approximation, fitting using Equation 7.10
achieved an R2 value of 0.989, see Fig. 16. The application of this approxi-
mation is dependent on the value of φ determined during equation fitting. As
the value of φ reduces the error associated with this approximation increases.
The value of φ here was sufficiently high that the error in the approximation
had no significant effect on the goodness of fit.
176
1.8
Sr = 100%
1.6 Constant e
1.4
Void Ratio, e
Constant ω
1.2
0.8
0.6
Constant σ
0.4
0 0.8
0.6
50 0.4
0.2
100 0
Sr = 100% for Gs = 2.68
er0 = 0.68, b = 0.10, ɸ = 45.0, m = 3.06 − R2 = 0.997
Figure 7.12: e − ω − σ equation surface fit to data from Tripathy et al. (2002)
Soil A
100
Sr = 100% for Gs = 2.68
90
80
0.5
70
0.7
60
50
1.0
40 1.1
30
1.3
20
1.5
10
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Figure 7.13: e − ω − σ equation constant e contours fit to data from Tripathy

et al. (2002) Soil A
178
2
Sr = 100% for Gs = 2.68
1.8
6.25
1.6
12.5
1.4
25
1.2
50
1
100
0.8
0.6
0.4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Figure 7.14: e − ω − σ equation constant σ contours fit to data from Tripathy

2
Sr = 100% for Gs = 2.68
Constant ω
1.8
0.5
1.6
1.4 0.4
1.2
0.3
1
0.8 0.2
0.6
0.1
0.0
0.4
0 20 40 60 80 100
Figure 7.15: e − ω − σ equation constant ω contours fit to data from Tripathy

3
2.5
σ = 6.25
σ = 10 kPa
2
σ = 20 kPa
σ = 30 kPa
σ
(∂ e/∂ ω(
σ = 40 kPa
1.5 σ↑ σ = 50 kPa
σ = 60 kPa
σ = 70 kPa
σ = 80 kPa
1
σ = 90 kPa
σ = 100 kPa
0.5
0
0 0.1 0.2 0.3 0.4 0.5
Gravimetric water content, ω (proportion)
0
ω↓
-1 ω= 0
ω = 0.325
ω = 0.025
ω = 0.35
ω = 0.05
ω = 0.375
-2 ω = 0.075
(1/kPa)
ω = 0.4
ω = 0.1
ω = 0.425
ω = 0.125
ω = 0.45
-3 ω = 0.15
ω
ω = 0.475
(∂ e/∂ σ(
ω = 0.175
ω = 0.5
ω = 0.2
ω = 0.525
-4 ω = 0.225
ω = 0.55
ω = 0.25
ω = 0.575
ω = 0.275
ω = 0.575
-5 ω = 0.3
-6
0 10 20 30 40 50 60 70 80 90 100
σ (kPa)
Figure 7.16: Partial derivatives of surface fit to data from Tripathy et al. (2002) Soil A. (a) Partial derivatives with respect to ω at
a range of σ’s (b) Partial derivatives with respect σ to at a range of ω’s
180
7.5 Discussion
The e − ω equation proposed here was fit to 20 sets of published shrinkage
curve data with the use of a single equation as shown above. In all cases the
goodness of fit, as determined by R2 , was above 0.99.
The proposed e − ω equation was able to model the entire shrinkage curve
using a single equation. The use of a single equation allows the transition
points between shrinkage zones and other characteristic features of the shrink-
age curve to be determined mathematically. This equation is flexible, being
able to model all configurations of the shrinkage curve. Soils which contain all
three zones of soil behaviour, those which lack a structural change zone and
those which lack a residual change zone, can all be modelled.
Unlike other existing equations, the proposed e−ω equation directly relates
features of the shrinkage curve to fitting parameters in the equation. This
requires the use of five fitting parameters to account for the transition from
the residual change zone to the proportional change zone, the slope of the
proportional change zone, the transition from the proportional change zone to
the structural change zone, the rate of change at each transition point and the
offset of the shrinkage curve from the saturation line. The relationship of these
parameters to soil properties is discussed in section 7.5.1.
The direct relationship between features of the shrinkage curve to fitting
parameters in the equation also allowed the equation to be extended to account
for the effect of net stress. The e − ω − σ equation proposed here was able to
achieve fit data published by (Tripathy et al., 2002) with the use of a single
equation, achieving an R2 above 0.997.
It should be noted that the equations proposed here are analytically dif-
ferentiable in explicit form. The derivatives of Equation 1 for both e − ω
and e − ω − σ forms are given in Appendix A. These derivatives are commonly
needed in incremental constitutive relations representing soil displacement due
to moisture and net stress change.
7.5.1 Theoretical understanding of fitting parameters in

the e − ω equation
For the purposes of this section, soil behaviour will be described in terms of
reducing water content from a saturated state.
Structural change is associated with the degree of macro porosity present in
the soil. As the water content of the sample decreases, the macro pores drain,
182
water preferentially to the micro pores (Braudeau et al., 2004). As the macro
pores drain air enters the sample, the result of which is a departure from the
saturation line. The volume change of a sample observed within this zone is
substantially less than the volume of water removed. This behaviour continues
until the swelling limit of the sample is reached. During drying, the swelling
limit is the water content where the macro pores have filled with air, and water
has not yet drained from micro pores. Alternatively during wetting, samples
experience most swelling below this limit, wetting beyond this limit results in
only macro pores being filled with water without much additional swelling.
In terms of the e − ω equation this point is described by fitting parameter a.
Based on this it is possible, where a soil exhibits a structural change zone, to
set fitting parameter a to the swelling limit.
During proportional volume change the soil behaviour is dependent upon

the material components of the soil sample. In an ideal case, volume decrease
is equal to the volume of water lost. In this situation, the slope of the shrinkage
curve in this zone will be equal to the value of the soils specific gravity, Gs , i.e.
parallel to the saturation line. Where water is lost from micro pores without
air entering into them (or without change in micro pore saturation) the ideal
case is observed, where air entry into micro pores does occur the slope of the
line will be less than Gs as the soil departs further from the saturation line.
Based on this, where a soil exhibits ideal behaviour, fitting parameter m can
be assigned a value of Gs . Physically, this behaviour is observed when clay
peds maintain full saturation during drying. Where non-ideal behaviour is
observed m must be determined from the data, and is expected to be less
than Gs . The proportional change zone continues until the shrinkage limit
of the sample is reached. Non-ideal behaviour is expected when there are
clay/slit/sand units in the matrix and associated pores desaturate successively
from largest to smallest.
Soil behaviour in the residual change zone is associated with desaturation

of micro pores due to air entry and material matrix becomes too stiff for soil
suction to produce shrinkage. Within this zone the decrease in sample volume
is again less than the volume of water lost due to the close contact between soil
particles. Fitting parameter b is associated with the shrinkage limit and it can
be assigned a value determined from the shrinkage limit. In an ideal soil, the
shrinkage limit of a soil sample occurs when the soil water content decreases
to a point when the soil shrinkage volume is no longer equal to the volume of
water lost, it is at this point that the proportional change zone transitions to
the residual change zone. Volume change in the residual change zone is less
than the volume of water lost as in this zone the micro pores in the soil are
desaturating. Desaturation occurs as the close contact between soil particles
prevents further volume reduction. Fitting parameter er is the intercept of
the shrinkage curve with the void ratio axis. er is therefore directly related to
the residual void ratio which is the minimum volume of voids in the sample
possible at the existing stress level.
There is no theoretical basis for the value of φ. The purpose of φ is to
control the rate of change of the curves gradient at the swelling and shrinkage
limits, fitting parameters a and b respectively. The value of φ may be related
to the presence of material components and pores that influence the transition
between different zones. In most cases the value of φ can be set to any value
above 200 with a high degree of confidence that this will not affect the fitting
of the curve. It should be noted that this is not always the case; where the
change in gradient of the curve occurs over a large range of water contents a
lower value for φ may be required to achieve an improved fit.
7.5.2 Theoretical understanding of fitting parameters in

the e − ω − σ equation
This section discusses the theoretical understanding of the extension to the
e − ω equation. Those fitting parameters not altered by the equation extension
to incorporate the effect of net stress are not discussed.
Increasing the net stress applied to a soil sample reduces the volume of
the voids contained in that sample. The result of this is the reduction in the
volume of water required to saturate the sample. This is clearly shown by the
consolidation behaviour of soil as shown by the 100% saturation line in Fig.
7.12. As the volume of voids is reduced by the increase in net stress conditions,
there is a reduction in the volume of water which can be held within micro
pores. This results in a decrease of the swelling limit of the sample, For this
reason, fitting parameter a must reduce as net stress increases to accommodate
this behaviour. In addition to the reduction in the volume of the micro pores
the volume of the macro pores is also reduced. The reduction in the volume
of the micro and macro pores reduces the residual void ratio of the soil and
therefore requires an associated reduction in fitting parameter er in order to
accommodate this behaviour. Reduction in the value of er as stress increases
results from structural rearrangement of soil particles in response to the higher
stresses.
184
7.6 Conclusions
This chapter has presented the development of a model for the relationship
between soil void ratio and soil water content. This model is extended to
incorporate the effect of net stress on the soil void ratio.
Application of the soil void ratio - soil water content model to literature
data shows that this model is able to fit a range of published shrinkage curve
data well, as measured by high R2 vales. The fitting parameters used are di-
rectly related to features of the relationship modelled. This direct relationship
means that they can be directly related to unsaturated and saturated soil me-
chanics theory and practise. The model proposed can be used to model soils
exhibiting all or only some of the zones of soil volume change behaviour.
All fitting parameters for the proposed model can be readily determined
manually from observation of the experimental data, while previous models
have required the use of trial and error or iterative methods (either manually
or through the use of optimisation software) for the majority of, if not all
fitting parameters.
The proposed soil void ratio - soil water content - net stress surface model
has also been shown to achieve an excellent fit to the literature data available.
It has also been shown that the derivatives of both models can be readily
derived. The partial derivatives of the surface model with respect to soil water
content and net stress are required for the modelling of expansive soil behaviour
as applicable to buried and surface structures with changes in soil water content
and net stress conditions.
Chapter 8
Modelling pipe-soil interaction
8.1 Introduction
This chapter presents a modelling approach to estimate the development of
flexural stresses in buried reticulation pipes due to the volumetric behaviour
of soil in response to changes in soil water content. Using the mechanical
properties of Australian cast iron pipe reported in Chapter 5 and the model
for the relationship between soil water content, net stress and soil void ratio
of environmentally stabilised soils reported in Chapter 7, this model will be
applied to the field study site (Chapter 4). Application of the model to the
field study site is used to validate the outputs of the model.
Section 8.2 discusses the options available for numerical modelling of soil
behaviour. Section 8.3 presents the theoretical framework used during the
development of the pipe-soil interaction model. Section 8.4 details the pipe-
soil interaction model. Section 8.5 applies this model to the instrumented
pipe segment and compares the estimations with the results obtained from the
field. Section 8.6 presents a parametric study of the model. Section 8.7 uses
the change in pipe flexural stresses estimated by the model for different soil
types to estimate the lifetime of these pipes.
8.2 Numerical modelling options

The response of unsaturated soils to changes in soil water content and stress
can be numerically modelled using a number of techniques, the most commonly
used of which are continuum models and Winkler spring models.
Winkler spring models represents soil in 2D as a series of separate springs
of known stiffness characteristics. This modelling approach assumes no direct
185
186
interaction between adjacent springs, i.e., adjacent springs cannot affect each
other via shear or volumetric affects. Where a number of springs are combined
either in series or in parallel the stiffness characteristics of these springs can
be combined to create a single representative spring.
Soil continuum models represent soil as a continuous medium where all
soil particles are interconnected. The introduction of the direct interaction
between soil particles significantly increases the complexity of models created
using this approach, introducing requirements for additional parameters and
equations to describe this interaction, e.g. the model presented by Alonso et al.
(1999). The increase in complexity also has a direct result of increasing the
computational time required to solve each step during modelling.
Despite less comprehensive approach offered by a Winkler spring model, it
has been adopted here due to the benefits of simplicity provided by this ap-
proach. The application of the more complex continuum models was deemed
not to be justifiable due to the large number of inputs required and the as-
sociated uncertainties of the current models for reactive soils. Rajani and
Tesfamariam (2004) used a Winkler spring model approach to develop an ana-
lytical model for behaviour of partially supported reticulation pipes for similar
reasons.
A Winkler spring approach where soil stress develops as the result of soil
expansion due to a change in water content has not previously been developed.
The basic principles for the application of the Winkler spring approach to
this situation is developed in this chapter. A simplified application of these
principles is also demonstrated.
8.3 Theoretical framework for modelling

It was shown in Chapter 7 that soil void ratio (e) can be written as a function
of soil water content (ω) and net stress (σ) . Figure 8.1 shows an idealised
representation of the surface relationship between e, ω and σ. The framework
below assumes soil volume change occurs in 1D only. 1D soil volume change
is referred to as soil height change for the remainder of this chapter.
The change in e can be written as a function of change in σ and change
in ω as shown in Equation 8.1. For the remainder of this chapter compressive
soil stress is considered to be positive.

∂e ∂e
de = dσ + dω (8.1)
∂σ ω ∂ω σ
CHAPTER 8. MODELLING PIPE-SOIL INTERACTION 187
e, soil void ratio
f(σ,ω)
(1+e0)
α (1+e0) Ec
σ, net stress
ω, soil water content
Figure 8.1: Idealised e − ω − σ surface
Equation 8.1 can be rewritten in terms of soil strain (ε) as shown in Equation
8.2 by dividing by 1 + e0 .
1
dε = .dσ − ασ .dω (8.2)
Ec
where
de
dε = − , (8.3)
1 + e0

∂e
1 ∂σ ω
=− , (8.4)
Ec 1 + e0

∂e
∂ω σ
ασ = , (8.5)
1 + e0
e0 is the initial void ratio, dε is the change in soil strain, E1c is the 1D con-
strained modulus of the soil at the state f (σ, ω) and ασ is the soil hydric
188
expansion coefficient at the state f (σ, ω).

Further to this, dε can be rewritten in terms of change in soil height re-
sulting in Equation 8.6. The effect of a change in ω and/or σ is now directly
related to a change in the height of the soil layer (in 1D).
dh 1
dε = − = .dσ − ασ .dω (8.6)
dz Ec
where dh is the change in height of a soil element and dz is the height of a soil
element.
The change in soil height can be calculated from Equation 8.6 via integra-
tion over the total layer height. This is shown in Equation 8.7.
Z H
1
∆h = − .dσ − ασ .dω dz (8.7)
0 Ec
where ∆h is the change in height of the soil layer in response to changes in σ

and/or ω, and H is the height of the soil layer.
Equation 8.7 can be rewritten to separate the effects of σ and ω as shown
in Equation 8.8.
Z H Z H
1
∆h = − .dσ.dz + ασ .dω.dz (8.8)
0 Ec 0
Z H Z H
1
.dσ.dz = ασ .dω.dz −∆h (8.9)
0 Ec 0
| {z }
∆hω
Following the assumption of 1D analysis and assuming dω = ∆ω for the soil

layers.
∆hω = ασ .∆ω.H (8.10)
where ∆hω is the change in soil height due to a change in ω assuming constant
σ.
Equation 8.9 shows that where the change in soil height and the change in
soil height possible solely as the result of a change in ω are known, the change
in σ can be calculated.
Again following the assumption of 1D analysis, dσ = ∆σ. Allowing Equa-
tion 8.9 to be rewritten as shown in Equation 8.11.
Z H
1
.∆σ.dz = ∆hω − ∆h (8.11)
0 Ec
The use of E1c in Equation 8.11 implies that the change in stress is constant
over the full layer height. This however is not the case. Elastic soil foundation
theory has been used to account for the attenuation of the stress imparted
to the soil directly beneath the pipe as depth increases (Das, 2007). Figure
8.2 shows a schematic representation of the attenuation of soil stress due to a
point load as depth increases. Equation 8.11 is modified to account for this,
resulting in Equation 8.12.
Z H
1
.If .∆σ.dz = ∆hω − ∆h (8.12)
0 Ec
where If is a factor which accounts for the attenuation of net stress due to
pipe restraint as the distance beneath the pipe increases and ∆σ represents
the stress at the calculation level.
Ground surface
Stress directly beneath pipe

due to the pipe
Increasing stress attenuation
Figure 8.2: Stress attenuation with depth
The solution to Equation 8.12, shown in Equation 8.13, is achieved using

the Steinbrenner solution for stress induced in an infinite layer due to a footing
of known length and width. Additionally a depth factor is also included to
account for the effect of burial depth (Das, 2007).

D (1 − ν)
∆hω − ∆h =∆σ 4. .Is .If
2 Ec
(8.13)
1
=∆σ.
ks
190
where ks is the soil spring modulus, D is the pipe diameter, ν is Poisson’s

ratio of the soil, Is is the shape factor (Steinbrenner, 1934) and If is the depth
factor (Fox, 1948). Is is calculated as shown in Equation 8.15.
ks is calculated as shown in Equation 8.14.
Ec
ks = (8.14)
2.D.(1 − ν).Is .If
1 − 2ν
Is = F1 + F2 (8.15)
1−ν
where
1
F1 = (A0 + A1 ) (8.16)
π
n0
F2 = tan−1 (A2 ) (8.17)
2π
√ √
(1 + m 02 + 1) m02 + n02
A0 = m0 .ln √ (8.18)
m0 (1 + m02 + n02 + 1)
√ √
(m0 + m02 + 1) 1 + n02
A1 = ln √ (8.19)
m0 + m02 + n02 + 1)
m0
A2 = √ (8.20)
n0 m02 + n02 + 1)
L
m0 = (8.21)
D
H
n0 = D (8.22)
2
H is the total soil height beneath the pipe and L is the distance between soil
springs, i.e., the length of modelled pipe section.
This theoretical framework presents a method for determining deformation
in a pipe-soil system resulting from a change in soil water content. Deformation
at the pipe level, ∆h, is dependent on pipe-soil interactions. A simplified
application of this theoretical framework to determine ∆h is presented in the
remainder of this chapter.
8.4 Model description

The response of the pipe-soil system to changes in soil water content is depen-
dent on the shrink/swell potential of the soil, the initial soil stress conditions
(state) and the interaction between the soil and the pipe. Change in soil
water content results in soil shrinkage or swelling. The amount of potential

shrink/swell can be modelled as linear model proportional to change in soil
water content.
In this model the change in soil water content is imposed onto the model
at each time step. The amount of soil shrink/swell determined at each time
step is applied to the model incrementally (i.e. ∆hω as in Equation 8.13). The
change in soil stress as a result of the soil shrink/swell is then determined.
The soil beneath the pipe is modelled as several layers according to the
geological formation and, differing soil water contents and stress conditions
as depth increases. Each soil layer is modelled as an individual spring. The
series of springs which represent the soil beneath the pipe are combined as a
representative spring.
The response of the soil to changes in soil water content and stress is
modelled using the surface equation presented in Chapter 7. This equation
is repeated below as Equation 8.23. The behaviour of the pipe and the pipe-
soil interaction are numerically modelled. It is important to note that this
surface is relevant only for environmentally stabilised soils, i.e., soils which
have experienced numerous cycles of wetting and drying such that its behaviour
has converged to a repeatable state. The pipe is modelled with linear elastic
material and complete contact between the pipe and soil is assumed. The
model described here is intended to only estimate the flexural stress developed
in the buried pipe. Pipe stress resulting from axial interactions are assumed
to remain constant. Where changes in the pipe-soil system result in changes
to the axial stress in the pipe, these changes must also be taken into account
where pipe failure is to be estimated.
e(ω) = f (ω, a) − f (ω, b) − f (0, a) + f (0, b) + er (8.23)
where;

m −1 1 2
f (x, y) = − φ(x − y)tan (φ(x − y)) − ln(1 + (φ(x − y)) ) (8.24)
φ.π 2

σ
er = αer ln 1 + + er0 (8.25)
σ0

σ
a = αa ln 1 + + a0 (8.26)
σ0
ω is the soil water content, e is the soil void ratio, σ is the soil net stress, σ0
192
is the nominal net stress, b, αer , er0 , αa , a0 , and m are fitting parameters.
The response of the pipe-soil system to changes in soil water content deter-
mined in five steps using a decoupled approach. Figure 8.3 shows a flow chart
of the modelling procedure.
Determine
initial Step 1
conditions
Determine soil
equilibrium stress σi+1
and deformation Δh i+1
Step 5
Step 4
Determine soil hydic Calculate volume change
expansion coefficient at constant stress Δhω
ασ,i and soil stiffness k s,i
Step 2
Step 3
Increment Soil
water content - ωi+1 i - time step
Figure 8.3: Flowchart of modelling procedure
The calculation of ks , ασ and hω are detailed in Sections 8.4.1, 8.4.2 and

8.4.3 respectively.
As the pipe-soil system is being modelled over time and using a series of
layers to represent the soil beneath the pipe, the subscripts i and j are added
to variables as necessary to identify the time step and layer respectively. For
example where the soil water content is described as ωi,j , this refers to the
soil water content in the ith time step, at the j th soil layer. Where a value
for subscript j is not included, this indicates consistent application to all soil
layers unless explicitly noted otherwise.
8.4.1 Calculation of soil stiffness, ks
ks is calculated in a two step procedure from the partial derivative of Equation

8.23 with respect to σ (Equation 8.27). The partial derivative is used to
calculate Ec as shown in Equation 8.4.

∂e αa .m −1 σ
=− .tan −φ αa .ln 1 + + a0
∂σ ω π.σ0 1 + σσ0 σ0

αa .m −1 σ
+ .tan φ ω − αa .ln 1 + + a0 (8.27)
π.σ0 1 + σ0σ σ0
αer
+
σ0 1 + σσ0
ks is then calculated using Equation 8.14. Investigation into the change in

ks over the range of soil water contents and net stresses to be used during
modelling found little change in ks . To simplify the numerical modelling a the
value of ks is calculated for each soil layer at initial conditions and then kept
constant. The values of ks calculated for each layer are combined to produce
a value to represent the full soil depth beneath the pipe using Equation 8.28,
as shown in Figure 8.4.
n
1 X 1
= (8.28)
ks,rep k
j=1 s,j
where ks,rep is the representative stiffness for the full soil depth, ks , j is the ks
of layer j and n is the number of soil layers beneath the pipe.
Soil layer 1
k s,1
Soil layer 2
k s,2
Representative
soil layer, k s,rep
Soil layer 3
k s,3
Soil layer 4
k s,4
Figure 8.4: Combination of soil layer to single representative spring

194
8.4.2 Calculation of soil hydric expansion coefficient, ασ

ασ is calculated from the partial derivative of Equation 8.23 with respect to
ω (Equation 8.29). The partial derivative is used to calculate ασ as shown in
Equation 8.5. ασ is calculated for each soil layer used in modelling. These
values are used to calculate the constant stress height change.

∂e m −1 σ m
= − tan (φ(ω − (αa .ln 1 + + a0 ))) + tan−1 (φ(ω − b))
∂ω σ π σ0 π
(8.29)
8.4.3 Calculation of change in soil height at constant

stress, hω
hω is calculated at each time step. The value of ασ is calculated using the σ
and ω conditions at the start of the time step as described above. At initial
modelling conditions. the existing stress within the system developed as a
result of the presence of the pipe is unknown. For this reason the initial
soil stress is assumed to be equal to the overburden stress, but a suitable
environmental condition such as wet/dry cycles could also have been chosen
to approximate this condition.
Unrestrained volume change is calculated as shown in Equation 8.30. Equa-
tion 8.30 is a modified version of Equation 8.9, where a field factor αf has been
added. αf has been included to account for the possibility of cracking and/or
for soil volume change occurring in multiple directions reducing the soil height
change, i.e. unrestrained or partial restrained behaviour. This factor is used
in residential slab and footing design in reactive soils (Standards Australia,
1996). The unrestrained volume change is calculated for each layer beneath
the pipe at each time step and summed.
n
X 1
∆hω,i = × ασ,ij × (ω(i+1)j − ωij ) × hij (8.30)
j=1
α f
where αf is the field factor (αf = 3 for unrestrained behaviour (Fityus et al.,
1998)).
8.4.4 Numerical Winkler model

A numerical model is used to determine the equilibrium conditions of the
pipe-soil system at each time step. The pipe-soil system is modelled as a
beam on a nonlinear Winkler foundation. A schematic representation of the

model is shown in Figure 8.5. The pipe is assumed to act as a continuous
unjointed cylinder, and represented is as an elastic beam element with either
fixed or pinned end support conditions. Selectable end support conditions
were included as Chan (2008) reported that end support conditions for buried
pipes as being between these two conditions. The end support conditions were
located at the interface between soil beneath pervious and impervious surfaces.
The soil above the pipe is represented as a uniformly distributed dead load.
Equilibrium is determined by supplying the numerical model with hω and ks .
Ground surface
Soil UDL
X X
Representative
soil layer
2% of pipe
length
Rock
X Pipe end conditions - both fixed and pinned

conditions investigared
Figure 8.5: Schematic of spring model
8.5 Application and validation of the model

In order to validate the model presented in this chapter it was applied to the
field study site detailed in Chapter 4. The mechanical properties of the pipe
used in modelling are given in Section 8.5.1, the e − ω − σ soil surface used to
196
represent the soil at the field study site is given in Section 8.5.2 and detail on
the modelling software used is given in Section 8.5.3.
8.5.1 Pipe mechanical properties

The mechanical properties of pipes similar to that instrumented at the field
study site were determined in Chapter 5. The pipe properties and dimensions
used in modelling are given in Table 8.1.
Table 8.1: Properties and dimensions of modelled cast iron pipe

Property Value Unit
Elastic modulus 150 GPa
Outside diameter 126 mm
Inside diameter 110 mm
Pipe length 20 m
8.5.2 Field site e − ω − σ soil surface

Application of the model to the field site required Equation 8.23 be fitted to
represent the soil beneath the pipe. Fitting was achieved using the results from
consolidation tests (Chan, InPrep), soil specific gravity, soil classification, and
soil water content and height change measurements. Soil bore hole analysis
(Chapter 4) found that the soil at the field study site was of a single type
beneath the pipe. Therefore, this soil could be represented using a single
surface. Multiple surfaces would be required where multiple soil types were
identified.
Initial fitting was undertaken using the results from consolidation tests.
The slope of the proportional shrinkage zone (m) was fixed to equal Gs . This
was believed to be a suitable figure as classification of the soil and the high
percentage of clay particles in the soil (' 60%).
Data collected using the neutron probe showed that soil water content
below the pipe was greater than 30% at all times. Based on this observation
the shrinkage limit (b) was set to 0.1. This value was chosen as it is low enough
to ensure that it did not affect calculations at the soil water contents measured
at the field study site. Whilst not necessarily the correct figure it was deemed
as a reasonable value.
As no detailed shrinkage curve data were available, the value of φ was set
to 500 to ensure a rapid transition between shrinkage zones. The soil water
content data also indicated that the soil was close to saturation. As a result,
the value of the swelling limit, a, selected would have a significant affect on
modelling. In order to determine if a must be determined for modelling, the
value of a was first set to a value above saturation. When a is set to a value
above saturation it no longer influences the surface, removing the structural
change zone.
The validity of setting a to a value above saturation was checked by ap-
plication of the model to estimate unrestrained soil height change away from
the pipe. For application of the model to estimate the unrestrained soil height
change the value of αf was set to 1. Setting αf to 1 imposed the assumption
that soil volume change was restrained to 1D height change. The result of this
analysis were compared with the results from the rod extensometer.
Figure 8.6 shows the unrestrained height change estimated by the model
and the field data from the rod extensometer. It can be seen that a field factor
of 1 over estimates soil height change, indicating that a field factor is needed
to more accurately represent field behaviour. A range of field factors were
applied, with a field factor of 1.75 found to produce model estimations which
closely follow the field data.
24
Unrestrained height change (mm)
20
16
12
-4
01/02/2009 01/06/2009 01/10/2009 01/02/2010 01/06/2010
Figure 8.6: Comparison of estimated unrestrained height change and rod ex-
tensometer measurements
Figure 8.7 shows the fitted surface with consolidation test. Lines of constant
e, ω and σ are also shown.
198
Consolidation test data
1.6
1.4
1.2
1
Gs
0.8
0.6
0.4
0.2
0
20 0.5
40 0.4
0.3
60 0.2
0.1
80 0
Figure 8.7: e − ω − σ surface fit for the Altona North field study site
The fitting and validation of the surface for the field study site allowed the
response of the pipe to changes in soil water content to be modelled. Modelling
has been undertaken using both fixed and pinned end support conditions. The
results of the modelling for fixed and pinned end support conditions are shown
in Sections 8.5.4 and 8.5.5 respectively. It should be noted that as a result
of data limitations (see Chapter 4 for details) modelling results can only be
compared with field data for the period between the 3rd of February 2009 and
the 3rd of August 2009. The results of modelling results are presented after
this time until the 17th of March 2010 without comparison data. The model
estimations were compared to the change in flexural stresses determined for
the instrumented pipe. The comparison is based on change in flexural stress
only. As was noted in Chapter 4 it is not possible to determine the existing

flexural stress in the pipe, i.e., the residual stress was not known, only the
change in flexural stress can be measured.
The soil water content measured at the field site are shown in Figure 8.8.
Soil water content data were available from the field study site at four depths,
approximately 200 mm apart vertically, from pipe depth. For the purposes of
modelling each position was associated with a single soil layer.
Based on this, the soil beneath the pipe at the field study site was repre-
sented using a series of soil layers 200 mm in height. The distance between
the pipe and bedrock at the field study site was 1.2 m, resulting in six layers.
As water content data were only available for four of the six layers the water
content for the bottom two layers is set to that measured for the fourth layer.
0.45
0.425
Gravimetric soil water content
0.40
Layer 1
(800 mm - 1000 mm)
0.375 Layer 2
(1000 mm - 1200 mm)
Layer 3
(1200 mm - 1400 mm)
Layers 4-6
0.35 (1400 mm - 2000 mm)
0.325
0.30
01/02/2009 01/05/2009 01/08/2009 01/11/2009 01/02/2009 01/05/2010 01/08/2010
Figure 8.8: Soil water content data from the Altona North field study site
8.5.3 Modelling software

The numerical Winkler model described in Section 8.4.4 was developed using
the OpenSEES software package (McKenna et al., 2000). OpenSEES is a finite
element package where system equilibrium is determined when unbalanced
system energy falls within a specified tolerance, set as 1.0 × 10−07 for this
200
model. The OpenSEES iterative solution algorithm uses Newton-Line search

method.
The pipe is modelled as a fibre section with linear elastic material proper-
ties. The soil beneath the pipe is modelled as non-linear springs evenly spaced
along the pipe length. The distance between soil springs (i.e. L) is equal to 2%
of the total length of the pipe being analysed which is 20 m. For the purpose
of this analysis, pipe flexural stress is defined as positive when the crown is in
tension relative to the invert at that location.
At each time step hω is applied at the bottom node of each soil spring
incrementally. hω is applied in a number of steps with each increment no
larger than 0.1 mm. The same soil displacement is assumed to occur along the
length of the modelled pipe as no data were available to indicate water content
variation along its length.
8.5.4 Modelling results for fixed end support conditions

This section presents the results of the modelling using fixed end support
conditions. Figure 8.9 shows the results for pipe deformation and the flexural
stress developed in the pipe as a result of soil volume change. An initial check of
the pipe deformation shows that the modelled pipe deformation closely follows
the effect of soil height change as a result of soil water content change. However,
inspection of the pipe flexural stresses developed in the pipe, specifically at the
ends are much higher than those expected. The tensile strength of cast iron,
as determined in Chapter 5, had a maximum value of approximately 250 M P a
and so pipe failure would be expected to occur at the stresses estimated by
the model.
For this reason high stresses are not considered to be an accurate esti-
mation. Examination of the stresses developed in the soil beneath the pipe
revealed high stresses developing at the ends during swelling. Comparison of
the change in pipe flexural stress determined from the field study site with the
change in the maximum pipe flexural stress estimated by the model is shown
in Figure 8.10. It can be seen that the trends of the modelled and field data
are similar, but the modelled stresses are much larger than those measured in
the field.
The high stresses developing in both the soil and pipe occurs as the result of
the fixed end conditions imposed during modelling. The fixed end conditions
do not allow vertical rotation of the pipe to occur at the ends, resulting the
development of a high bending moment, and therefore high stress.
14
12
01/01/2009
10
03/02/2009
10/03/2009
16/06/2009
Deformation (mm)
8
28/07/2009
01/09/2009
6 15/09/2009
13/10/2009
25/11/2009
4 28/01/2010
18/03/2010
2 16/04/2010
24/05/2010
−2 3
0 5 10 15 20 x 10
Along the Pipe (mm)
(a) Pipe deformation
100
50
0 01/01/2009
03/02/2009
10/03/2009
− 50 16/06/2009
28/07/2009
Stress (MPa)
01/09/2009
−100 15/09/2009
13/10/2009
25/11/2009
−150 28/01/2010
18/03/2010
16/04/2010
− 200
24/05/2010
− 250
− 300 3
0 5 10 15 20 x 10
Along the Pipe (mm)
(b) Pipe flexural stress
Figure 8.9: Modelling estimations for fixed end support conditions

202
20 100
Observed flexural stress
Modelled flexural stress - fixed end support
10 50
Change in flexural stress (MPa) - MODELLED

Change in flexural stress (MPa) - OBSERVED
0 0
-10 -50
-20 -100
-30 -150
-40 -200
-50 -250
-60 -300
01/02/2009 01/06/2009 01/10/2009 01/02/2009 01/06/2010
Figure 8.10: Comparison of the change in pipe flexural stress estimated for
fixed end support conditions and measurements from the Altona North field
study site
8.5.5 Modelling results for pinned end support condi-

tions
This section presents the results of the modelling using pinned end support
conditions. Figure 8.11 shows the results for pipe deformation and the flexural
stress developed in the pipe as a result of soil volume change. An initial check of
the pipe deformation shows that the modelled pipe deformation closely follows
the effect of soil height change as a result of soil water content change.
Figure 8.12 shows a comparison of the change in pipe flexural stress deter-
mined from the field study site, with the change in the maximum pipe flexural
stress estimated by the model. It can be seen that the model estimations do
not follow the trend seen in the results from the field study site. Further exam-
ination of the model estimations found, the magnitude of the values estimated
by the model to be similar to those from the field, but opposite in sign, refereed
to as inverse estimation for the remainder of this chapter.
The pinned end support condition allows free vertical rotation of the pipe
to occur at the ends. Examination of modelling results found that the inverse
estimation occurred due to this free rotation.
14
12
01/01/2009
10 03/02/2009
10/03/2009
16/06/2009
Deformation (mm)
8
28/07/2009
01/09/2009
6 15/09/2009
13/10/2009
25/11/2009
4 28/01/2010
18/03/2010
16/04/2010
2
24/05/2010
−2 3
0 5 10 15 20 x 10
Along the Pipe (mm)
(a) Pipe deformation
80
70
60 01/01/2009
03/02/2009
50 10/03/2009
16/06/2009
28/07/2009
Stress (MPa)
40 01/09/2009
15/09/2009
30 13/10/2009
25/11/2009
28/01/2010
20 18/03/2010
16/04/2010
10 24/05/2010
−10 3
0 5 10 15 20 x 10
Along the Pipe (mm)
(b) Pipe flexural stress
Figure 8.11: Modelling estimations for pinned end support conditions

204
80 80
Observed flexural stress
Modelled flexural stress - pinned end support
60 60
Change in flexural stress (MPa) - MODELLED

Change in flexural stress (MPa) - OBSERVED
40 40
20 20
0 0
-20 -20
-40 -40
01/02/2009 01/06/2009 01/10/2009 01/02/2009 01/06/2010
Figure 8.12: Comparison of the change in pipe flexural stress estimated for
pinned end support conditions and measurements from the Altona North field
study site
8.5.6 Averaged modelling results

Sections 8.5.4 and 8.5.5 estimated pipe flexural stresses for fixed and pinned
end support conditions respectively. The use of fixed end support conditions
in modelling resulted in a over estimation of pipe flexural stress, whilst the use
of pinned end support conditions resulted in inverse estimations. These results
indicate that in the field the end support conditions exist between these two
extremes, this conclusion is supported by Chan (2008).
To investigate this, the results for modelling using fixed and pinned end
support conditions were averaged and compared to the results from the field
study site. A comparison of the averaged results and the change in pipe flexural
stress determined at the field study is shown in Figure 8.13. The averaged
estimations closely follow the trend observed in the field.
This result indicates that in the field a situation of limited pipe rotation
exists at the interface between soil beneath pervious and impervious surfaces.
40
Observed pipe flexural stress
Modelled flexural stress - averaged
20
0
Change in flexural stress (MPa)
-20
-40
-60
-80
-100
-120
01/02/2009 01/06/2009 01/10/2009 01/02/2009 01/06/2010
Figure 8.13: Comparison of the change in the pipe flexural stresses averaged
between pinned and fixed end support conditions, and measurements from the
Altona North field study site
8.6 Parametric study of the model

This section presents a parametric study of the model to examine the effect of
changing ks and ασ , ∆ω and the uniformly distributed dead load. The average
of modelling results for fixed and pinned end support conditions were used in
this section as it was shown in the previous section that these results closely
follow the trend observed at the field study site. Although the parametric
study can be defined in non-dimensional terms, dimensional values have been
used here to show the direct influence of the changed values. This is applicable
here as the geometry of the scenario remains constant.
Based on the resulted obtained from the field study site used in Section
8.5, a base line soil water content cycle has been created for six soil layers.
Layer 1 is directly beneath the pipe with layer 6 being directly above bedrock.
Layers 4 - 6 have the same soil water content cycle. The soil water content
cycles for each layer are shown in Figure 8.14. The magnitude of the base line
soil water content cycles was created to represent the extremes of soil water
content occurring over the year. For the remainder of this chapter, the base
line soil water content cycle will be referred to as the base line cycle, and the
magnitude of the base line cycle will be referred to as the base line magnitude.
206
The minimum soil water content has been set to occur in March, in line
with the observed peak in pipe failure numbers seen in Chapter 3. However,
no attempt has been made to correlate the changes in the soil water content
to specific months.
0.50
0.45
0.40
Gravimetric soil water content
0.35
Layer 1
0.30 (800 mm - 1000 mm)
Layer 2
(1000 mm - 1200 mm)
0.25 Layer 3
(1200 mm - 1400 mm)
Layers 4-6
0.20 (1400 mm - 2000 mm)
0.15
0.10
0.05
0
01/01/2010 01/01/2011 01/01/2012 01/01/2013 01/01/2014 01/01/2015
Figure 8.14: Base line soil water content cycle
The parametric study was conducted using three soil types to represent
very expansive, expansive and stable soils. A summary of the ks and ασ values
used in this parametric study are shown in Table 8.2. As the data required to
create full surfaces for all three soil types was unavailable, values for ks and ασ
were determined separately using values obtained for soils deemed to belong
to each soil type.
ks values were calculated following the process described in Section 8.4.1
from Ec values representative of these soil types. Ec values were calculated
from the Young’s modulus using Equation 8.31 (Fredlund and Rahardjo, 1993).
E values for stable soil were obtained for sand from Kulhawy and Mayne
(1990). E values for expansive soils were obtained for clay from Costa (2009).
As soil water content has a strong affect on soil behaviour, where possible
values were selected for soil water contents close to the range observed at the
field study site (35% / ω / 44%)
E(1 − ν)
Ec = (8.31)
(1 + ν)(1 − 2ν)
where E is the soil’s Young’s modulus and ν is the soil’s Poisson’s ratio.
Values for ασ were obtained for very expansive and expansive soil from the
shrinkage curves presented in Chapter 7. ασ for stable soil was set at a nominal
value of 0.01, as by definition stable soils are not expansive. For the purpose
of this study the values of ks and ασ were assumed to remain constant over
the applied range of soil water contents.
The values for ks and ασ used to represent very expansive, expansive and
stable soils as given in Table 8.2.
Table 8.2: Values of ks and ασ for different soil types
Shrink/swell
Soil Type ks ασ
potential
Very Expansive High 0.11 1.29
Expansive Moderate 0.16 0.60
Stable Low 0.28 0.01
8.6.1 The effect of changing soil stiffness and hydric ex-

pansion coefficient
This section uses values of ks and ασ of very expansive, expansive and stable
soils shown in Table 8.2 to investigate the effect of soil type on modelling. Soil
water content change is imposed using three levels of soil water content change;
the base line magnitude, 50% of the base line magnitude and 200% of the base
line magnitude. Figures 8.16 and 8.15 show the change in pipe flexural stress
resulting from the change in soil water content associated with the ks and ασ
for each soil type respectively.
It can be seen in Figures 8.15 and 8.16 that the highest flexural stresses
develops due to the action of very expansive soil. The flexural stresses de-
veloped due to the action of expansive soil were approximately 50% of those
developed in very expansive soil for each level of change in soil water content.
The stable soil has an negligible affect on the flexural stresses developed in the
buried pipe.
These results were in-line with expectations based on the investigation
presented in Chapter 3. Pipes in very expansive soils were seen to have the
highest failure rate, with the failure rate reducing as soil shrink/swell potential
reduced.
208
160
100% Base line ∆ ω
120
∆ σ (MPa)
80
40
0
0.10 0.14 0.18 0.22 0.26 0.3
ks (MPa)
Figure 8.15: Change in pipe flexural stress for three soil types and water
content changes against ks .
It can be seen in Table 8.2 that as soil reactivity increases, ασ increases

and ks decreases. As ks decreases the soil requires less stress to deform it,
which indicates that as the shrink/swell potential of soil increases soil becomes
softer and easier to compress, and therefore reducing its ability to induce stress
in a buried pipe. The increase in ασ as soil shrink/swell potential increases
results in great volume change as soil water content changes, increasing the
deformation imposed on a buried pipe. The results shown in Figures 8.16 and
8.15 show that whilst ks decreases as its shrink/swell potential increases, the
effect of this decrease is more than offset by the effect of the increase in ασ .
Therefore the ασ of a soil is the more significant factor in determining the
stresses imposed on a buried reticulation pipe by that soil.
8.6.2 Effect of soil water content change

The effect of soil water content change on the development of pipe flexural
stress was investigated with reference to the base line cycle shown in Figure
8.14. The change in soil water content was set to 50%, 100% and 200% of the
160
100% Base line ∆ ω
120
∆ σ (MPa)
80
40
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
ασ
Figure 8.16: Change in pipe flexural stress for three soil types and water
content changes against ασ .
base line magnitude. The effect of soil water content change was investigated
for all three soil types.
Figure 8.17 shows the effect of the base line cycle, 50% of the base line
magnitude and 200% of the base line magnitude on the change in flexural
stress developed in a buried pipe in very expansive soils, expansive soils and
stable soils.
It can be seen in Figure 8.17 that as the change in soil water content
increases the change in flexural stress increases in a linear manner. When the
change in soil water content is reduced from the base line magnitude to 50% of
the base line magnitude the change in flexural stresses reduces by around 50%
also. Similarly when the change in soil water content is increased from the base
line magnitude to 200% of the base line magnitude the change flexural stresses
increases by around 200% also. The response was observed for all three soil
type.
This change can be seen to be significant for very expansive and expansive
soil, but negligible for stable soils. Therefore the change in soil water content is
210
160
120
∆ σ (MPa)
80
40
0
50 100 150 200
% base line magnitude Δ ω (VE soil)
Average base line magnitude Δ ω = 1.6%
Figure 8.17: Change in pipe flexural stress for 50%, 100% and 200% of the
base line magnitude
a significant factor in determining the stresses imposed on a buried reticulation

pipe.
The significance of the change in soil water content has an interactive affect
with ασ as a level of stress developed in a buried reticulation pipe is dependant
on both factors. A large change in the soil water content will have little affect
where the pipe is buried in a soil with a low shrink swell potential, and a pipe
buried in a soil with a high shrink swell potential will not be affected if there
is little change in the soil water content. Conversely a large change in the
soil water content of a soil with a high shrink swell potential will result in a
significant change in the flexural stress of that pipe.
8.6.3 Effect of changing the uniformly distributed dead

load
The uniformly distributed dead load applied to the pipe is the result of soil
overburden. The uniformly distributed dead load was altered in the model by
increasing the unit weight of the soil above the pipe.

It was found that change in the uniformly distributed dead load has a
negligible affect on the change in flexural stress. This is likely due to the
contact between the buried pipe and its surrounding soil. In a situation where
the soil shrinks to such an extent that the soil separates from the pipe it
is expected that the uniformly distributed dead load will strongly affect the
flexural stress in the buried pipe.
8.7 Pipe lifetime estimation

This section presents an estimation of the lifetime of a cast iron pipe buried in
very expansive, expansive and stable soils. The capacity of the modelled pipe
is reduced using both methods presented in Chapter 5. In both methods the
capacity of the pipe material remains constant, as does the pipe deformation.
In the first method pipe wall thickness reduces uniformly causing a decrease
in the second moment of area, resulting in an increase in stress and a net
section collapse failure mode. In the second method the applied stress intensity
factor (KI ) increases with in pit depth, resulting in a crack growth failure
mode. In both methods the contribution of the graphite structure retained
after graphitisation is assumed to be negligible, i.e. graphitised areas have no
residual strength.
Pipe loading was isolated to the change in flexure due to soil volume change
using the ks and ασ of the very expansive soil, and the base line cycle. Pipe
deterioration rate, wall thickness reduction or pit growth, was calculated based
on the corrosion rate. The corrosion rate calculated from Equation 8.32, using
the fitting parameters shown in Table 8.3 (Rajani et al., 2000). These fitting
parameters were determined by Rajani et al. (2000) using measurements of
maximum and average pit depth, and pipe age to determine the maximum
and average corrosion rates respectively. The maximum pit depths were deter-
mined using point measurements. Whilst the total material lost was used to
determine the average reduction in wall thickness, i.e., the average pit depth.
Based on this the pipe wall thickness reduction was calculated using the aver-
age corrosion rate, whilst pit depth increase was calculated using the maximum
corrosion rate.
cr = a + b.c.e−c.t (8.32)
where cr is the corrosion rate, a, b and c are model constants, and t is time
212
Table 8.3: Constants for exponential corrosion model (Rajani et al., 2000)
Corrosion Rate a b c
(mm/year) (mm/year) (mm/year) (/year)
Average 0.0042 1.95 0.058
Maximum 0.0125 5.85 0.058
(years).
Figure 8.18 shows the increase in pipe flexural stress over time due to the
action of general corrosion. It can be seen that failure by net section collapse
is not estimated to occur within 200 years of pipe installation. Figure 8.19
shows the increase in pipe flexural stress over time due to the action of pit
corrosion. It can be seen that failure by crack growth is estimated to occur
after 49 years in very expansive soil. The estimation of failures occurring
within 50 years of installation by crack growth due to flexural stresses, .i.e,
circumferential failures, is considered to be reasonable. Examination of the
historical data presented in Chapter 3 found records of circumferential failures
occurring pipes of this age and even younger.
200
Pipe stress VE soil
Pipe stress EX soil
180
Pipe stress ST soil
Pipe failure stress
160
140
120
Pipe stress (MPa)
100
80
60
40
20
0
0 20 40 60 80 100 120 140 160 180 200
Time (years)
Figure 8.18: Pipe lifetime estimation incorporating soil type and corrosion rate
for net section collapse failure
14
12
10
8
KI (MPa. m)
KI VE soil
KI EX soil
2
K ST soil
I
Cast Iron KIC
0
0 20 40 60 80 100 120 140 160 180 200
Time (years)
Figure 8.19: Pipe lifetime estimation incorporating soil type and corrosion rate
for crack growth failure
8.8 Conclusions
This chapter has presented a modelling approach for the estimation of flexural
stress development in buried reticulation pipes due to the soil volume change
that results from changes in soil water content. The pipe-soil interaction model
accounts for the effect of changing soil water content and the restraint imposed
on the soil by the presence of the reticulation pipe.
Application of the pipe-soil model to represent the field study site detailed
in Chapter 4 showed that neither the fixed nor pinned end support conditions
estimated good results, when the estimated stresses were compared to stress
data obtained from the site. The fixed end support condition was found to show
a similar shape trend to the observed data, but over estimated stress levels.
The pinned end support condition estimated stresses of similar magnitude of
the values to those from the field, but opposite in sign. Averaging of the values
estimated by these extremes found good agreement with the field data. This
indicated that in the field a situation of limited pipe rotation exists at the
interface between soil beneath pervious and impervious surfaces.
214
A parametric study of the model showed that flexural stress development

is strongly influenced by both soil hydric expansion coefficient and soil water
content change. The effect of the decrease in soil stiffness as the soil hydric
expansion coefficient increased, was found to be overshadowed by the effect of
the change in soil hydric expansion coefficient. The uniformly distributed load
had a negligible affect on flexural stress development.
The lifetime of pipes buried in very expansive, expansive and stable soils
due to flexure was estimated for both net section collapse and crack growth
failure modes. Failure was not estimated to occur as the result of net section
collapse within 200 years of installation. However, failure by crack growth was
estimated to occur within 50 years.
Chapter 9
Conclusions and
recommendations
9.1 Introduction
The number of failures in buried water reticulation pipe networks varies ac-
cording to season, peaking during mid summer to early autumn in Melbourne,
Australia. The magnitude of this peak also varies between years to such a de-
gree that the the peak is strongly evident in some years and scarcely observable
in others.
A lack of understanding of the causes of these variations means that it is
not possible to predict when or even if failures will peak at specific times within
the year. This lack of understanding reduces the effectiveness of management
expenditure.
This thesis has presented a systematic investigation into the causes of the
seasonal variations in the occurrence of failures in buried water reticulation
pipe networks.
The research presented in this thesis was undertaken in three major stages.
The conclusions from each of these stages are presented in Section 9.2 and
recommendations for future work are given in Section 9.3.
9.2 Conclusions
Stage one of this study was an investigation of historical pipe failure data from
two water authorities operating within the metropolitan area of Melbourne,
Australia. The data were investigated to identify the possible causes of the sea-
sonal peak in pipe failures. The occurrence of the seasonal peak was found to
215
216
be strongly affected by climatic conditions, pipe material type and surrounding

soil type. In particular, more failures occur during drier climatic conditions for
cast iron pipes buried in soils with a high shrink/swell potential The outcome
of this stage of the research was the hypothesis that the seasonal variation in
pipe failures resulting in a peak in pipe failures during mid summer to early
autumn in Melbourne, Australia occurs as a result of soil shrinkage.
Stage two was a detailed field study of an in-service pipe, its surrounding
soil and climatic conditions. The field study was undertaken on an in-service
cast iron reticulation pipe with a 100 mm nominal diameter buried in a soil
with high shrink/swell potential. A number of instruments were installed dur-
ing the field study to collect and record data including pipe strain, soil water
content and soil height change. Analysis of the results showed that between
mid summer and early autumn, positive flexural stress developed in the pipe,
with tensile stress at the crown and compressive stress at the invert. This
flexural stress was directly related to soil volume change, increasing simulta-
neously with soil shrinkage and reduction in soil water content. The increase
in pipe stress due to soil shrinkage is the cause of pipe failure, thus lending
strong support to the hypothesis that soil shrinkage is the cause of the seasonal
variation in pipe failures.
Stage three was the development of a model to predict the development
of flexural stresses in pipes resulting from the change in soil water content.
This model was based on the relationship between soil void ratio, water con-
tent and net stress. The partial derivatives of this relationship were used to
calculate the soil stiffness and the hydric expansion coefficient. These values
were used in conjunction with a Winkler spring numerical model, to determine
equilibrium conditions for a pipe-soil system subject to changes in soil water
content. The relationship between soil void ratio, water content and net stress
was represented by a new surface, developed using a novel technique which
allowed fitting parameters to be directly related to the characteristic features
of the relationships. The outputs of the model were compared to results ob-
tained from the field study site. It was found that the use of fixed and pinned
end support conditions failed to accurately predict the development of flexural
stress. Further investigation found that field conditions existed between these
two extremes, indicating that in the field a situation of limited pipe rotation
exists at the interface between soil beneath pervious and impervious surfaces.
Thus the research presented in this thesis has provided a clear understand-
ing of the causes of the seasonal variation in reticulation pipe failure numbers.
CHAPTER 9. CONCLUSIONS AND RECOMMENDATIONS 217
This understanding has enabled the development of a model to represent the

complex interaction between climatic conditions, soil, and the flexural stress
in buried pipes.
9.3 Recommendations for future work

To provide further understanding of the behaviour of buried reticulation pipes
in response to the effects of climatic conditions and different surrounding soils,
the research reported in this thesis could be extended in a number of ways.
9.3.1 Pipe joint characterisation

The numerical modelling undertaken in Stage three of this research assumed
that the buried pipe acted as a continuous unjointed cylinder. In reality the
pipe is a series of small pipe segments connected using bell and spigot type
joints. Laboratory work to characterise these joints to determine, available
rotation and load transfer between segments, is recommended to improve the
accuracy of future modelling.
Further work to develop appropriate numerical representations of these
joints is also recommended.
9.3.2 End support characterisation

During the modelling carried out during this research, fixed-fixed or pinned-
pinned end support conditions were applied to the interface between soil be-
neath pervious and impervious surfaces. Comparison of the results from mod-
elling to results obtained from the field study indicate that the true end support
condition is between these two extremes. Laboratory and field work to char-
acterise the end support conditions which exist at the interface between soil
beneath pervious and impervious surfaces is recommended.
Further work to develop appropriate numerical representations of this in-
terface is also recommended.
9.3.3 Validation of the e − ω − σ model

This thesis has presented a new model which relates soil stress, soil water
content and void ratio in a practical model which is not subject to the hysteresis
inherent in the use of matric suction. This model has the potential to replace
218
the current use of models which currently make use of relationships between
suction, soil stress and void ratio whilst assuming no hysteresis.
It is recommended that further work of this model be undertaken to validate
this model to enable its application in this manner.
9.3.4 Continuum modelling

This thesis has presented a simplified model for the behaviour of buried pipes
in reactive soils that incorporates the observed behaviour of reactive soils.
More sophisticated 3D continuum numerical modelling can be undertaken
to extend this modelling. This more sophisticated model would enable a more
detailed examination of pipe performance. Such modelling requires a general
numerical model for reactive soil to be developed. Accordingly the Winkler
model proposed in this thesis could be further refined if necessary.
9.3.5 Asset management application

This thesis has presented a model for the interaction between climatic condi-
tions, soil, and the flexural stress in buried pipes.
To allow the research presented in this thesis to be fully utilised by asset
managers, further work to fully develop the implications for industry is recom-
mended. Such work should focus on the development of processes to identify
assets likely to be subject to increased loadings as a result of the combined
action of climatic conditions and surrounding soil types.
Chapter 10
References
Al-Adeeb, A. M. and Matti, M. A., 1984. Leaching corrosion of asbestos
cement pipes. International Journal of Cement Composites and Lightweight
Concrete, 6(4): 233–240.
Alonso, E. E., Vaunat, J. and Gens, A., 1999. Modelling the mechanical
behaviour of expansive clays. Engineering geology, 54(1): 173–183.
ASTM International, 1999. ASTM E8M 04: Standard Test Methods for
Tension Testing of Metallic Materials [Metric].
ASTM International, 2003. ASTM D4546 One-dimensional swell or settlement

potential of cohesive soils.
ASTM International, 2006. ASTM E 399 06: Standard Test Method for
Linear-Elastic Plane-Strain Fracture Toughness Kic of Metallic Materials.
Atkinson, K., Whiter, J. T., Smith, P. A. and Mulheron, M., 2002. Failure of
small diameter cast iron pipes. Urban Water, 4(3): 263–271.
Baracos, A., Hurst, W. and Legget, R., 2010. Effects of physical environment
on cast-iron pipe. Journal AWWA, 47(12): 1195–1206.
Beech, M., 2007. Personal communication on usage era of pipe materials.
Blanchard, B. J., McFarland, M. J., Schmugge, T. J. and Rhoades, E., 1981.

Estimation of Soil Moisture with API Algorithms and Microwave Emission.
Journal of the American Water Resources Association, 17(5): 767–774.
Boxall, J. B., O’Hagan, A., Pooladsaz, S., Saul, A. J. and Unwin, D. M., 2007.
Estimation of burst rates in water distribution mains. Proceedings of the
Institution of Civil Engineers-Water Management, 160(2): 73–82.
219
220
Braudeau, E., Costantini, J., Bellier, G. and Colleuille, H., 1999. New device
and method for soil shrinkage curve measurement and characterization. Soil
Science Society of America Journal, 63: 525–535.
Braudeau, E., Frangi, J.-P. and Mohtar, R. H., 2004. Characterizing Nonrigid
Aggregated Soil-Water Medium Using its Shrinkage Curve. Soil Science
Society of America Journal, 68(2): 359–370.
Briaud, J.-L., Zhang, X. and Moon, S., 2003. Shrink Test–Water Content
Method for Shrink and Swell Predictions. Journal of Geotechnical and
Geoenvironmental Engineering, 129(7): 590–600.
Brooks, R. H. and Corey, A. T., 1964. Hydraulic properties of porous media.
Buisson, M. S. R. and Wheeler, S. J., 2000. Inclusion of hydraulic hysteresis in

a new elsto-plastic framework for unsaturated soils. Proceedings of Exper-
imental Evidence and Theoretical Approaches in Unsaturated Soils, edited
by Tarantino, A. and Mancuso, C. A.A. Balkema, Rotterdam, 109–119.
Burger, C. A. and Shackelford, C. D., 2001. Soil-Water Characteristic Curves

and Dual Porosity of Sand–Diatomaceous Earth Mixtures. Journal of
Geotechnical and Geoenvironmental Engineering, 127(9): 790–800.
Burn, S., Davis, P., Schiller, T., Tiganis, B., Tjandraatmadja, G., Cardy,
M., Gould, S., Sadler, P. and Whittle, A., 2005. Long-Term Performance
Prediction for PVC Pipes. AWWARF.
Caproco Corrosion Prevention Ltd, 1985. Underground corrosion of water

pipes in Canadian cities. Case: The city of Calgary. Technical report, Rep.
Prepared for CANMET.
Chan, D., 2008. Performance of water and gas pipes buried in reactive soil.
Master’s thesis.
Chan, D., InPrep. Study of pipe-soil-climate interaction of buried water and

gas pipes. Ph.D. thesis.
Chan, D., Kodikara, J., Gould, S., Ranjith, P., Choi, X. S. K. and Davis,
P., 2007. Data analysis and laboratory investigation of the behaviour of
pipes buried in reactive clay. Proceedings of 10th Australia New Zealand
Conference on Geomechanics - Common Ground 2007.
CHAPTER 10. REFERENCES 221
Chapuliot, S., 2000. Formulaire de KI pour les tubes comportant un dfaut

de surface semi-elliptique longitudinal ou circonfrentiel, interne ou externe.
Technical Report Rapport CEA-R-5900.
Chertkov, V., 2000. Modeling the pore structure and shrinkage curve of soil
clay matrix. Geoderma, 95(3-4): 215–246.
Chu, T. Y. and Mou, C. H., 1973. Volume change characteristics of expansive

soils determined by controlled suction test. Proceedings of 3rd International
Conference on Expansive Soils, volume 1. 177–185.
Ciottoni, A. S., 1985. Updating the New York City water system. Proceedings
of the specialty conference on infrastructure for urban growth. 69–77.
Clark, C. M., 1971. Expansive- soil effect on buried pipe. Journal AWWA,
63(7): 424–427.
Colin, R. and Baker, T., 1991. Application of fracture mechanics to the failure
of buried cast iron mains. Technical report, Transport and Road Research
Laboratory.
Constantine, A. and Darroch, J., 1993. Stochastic Models in Engineering

Technology and Management. Proceedings of Australia-Japan Workshop on
Stochastic Models in Engineering, Technology and Management, edited by
Osaki, S. and Murthy, D. Singapore: World Scientific, 86–95.
Constantine, A., Darroch, J. and Miller, R., 1996. Predicting Underground

Pipeline Failure. Water: 9–10.
Costa, S., 2009. Study of Desiccation Cracking and Fracture Properties of

Clay Soils. Ph.D. thesis.
Das, B. M., 2007. Principles of foundation engineering. Thomson, 6 edition.
Davis, P., Burn, S. and Gould, S., 2008a. Fracture prediction in tough
Polyethylene pipes using measured craze strength. Polymer Engineering
and Science, 48(5): 843–852.
Davis, P., Burn, S., Gould, S., Cardy, M., Tjandraatmadja, G. and Sadler, P.,
2007a. Long-Term Performance Prediction for PE Pipes. AWWARF.
Davis, P., Burn, S., Moglia, M. and Gould, S., 2007b. A physical probabilistic
model to predict failure rates in buried PVC pipelines. Reliability Engineer-
ing and System Safety, 92(9): 1258–1266.
222
Davis, P., De Silva, D. and Gould, S., 2009. Improving Asset Management of
Water Pipelines Through Understanding Materials Durability and Failure.
Proceedings of OzWater 09. AWA.
Davis, P., De Silva, D., Marlow, D., Moglia, M., Gould, S. and Burn, S., 2008b.
Failure prediction and optimal scheduling of replacements in asbestos cement
water pipes. Journal of Water Supply Research and Technology-Aqua, 57(4):
239–252.
Davis, P., Moglia, M., Burn, S. and Farlie, M., 2004. Estimating failure prob-
ability from condition assessment of critical cast iron water mains. Proceed-
ings of 6th National Australasian Society for Trenchless Technology Trench-
less Conference.
Dickey, G. L., 1990. Factors affecting neutron gauge calibration. Irigation and
Drainage. ASCE.
Doherty, J., 2003. PEST - Surface Water Utilities. Watermark Numerical

Computing and University of Idaho.
Dyt, K., 2009. Personal communication on pipe installation practise.
Fityus, S. G., Walsh, P. and Kleeman, P. W., 1998. The Influence of Climate
as expressed by the Thornthwaite Index on the Design Depth of Moisture
Change of Clay Soils in the Hunter Valley. Proceedings of Conference on
geotechnical engineering and engineering geology in the Hunter Valley. 261–
265.
Fleureau, J., Verbrugge, J., Huergo, P., Correia, A. and Kheirbek-Saoud, S.,
2002. Aspects of the behaviour of compacted clayey soils on drying and
wetting paths. Canadian Geotechnical Journal, 39(6): 1341–1357.
Fox, E. N., 1948. The mean elastic settlement of a uniformly loaded area at a
depth below the ground surface. 2130.
Fredlund, D. G. and Rahardjo, H., 1993. Soil mechanics for unsaturated soils.
John Wiley and Sons, Inc., USA.
Fredlund, D. G. and Xing, A., 1994. Equations for the soil-water characteristic
curve. Canadian geotechnical journal, 31(4): 521–532.
Fredlund, M. D., Wilson, G. W. and Fredlund, D. G., 2002. Representation

and estimation of the shrinkage curve. Proceedings of 3rd International Con-
ference on Unsaturated Soils, UNSAT 2002, volume 1. Taylor and Francis,
Rotterdam, The Netherlands, 145–150.
Gallage, C., Chan, D., Gould, S. and Kodikara, J. K., 2009. Stress-Strain
Development Of An In-Service Cast Iron Water Reticulation Pipe Buried In
Expansive Soil. Proceedings of OzWater 09. AWA.
Gitirana, J., de F. N. Gilson and Fredlund, D. G., 2004. Soil-Water Character-

istic Curve Equation with Independent Properties. Journal of Geotechnical
and Geoenvironmental Engineering, 130(2): 209–212.
Gorji-Bandpy, M. and Shateri, M., 2008. Analysis of Pipe Breaks in Urban

Water Distribution Network. GREATER MEKONG SUBREGION ACA-
DEMIC AND RESEARCH NETWORK, 14: 117.
Gould, S., Boulaire, F., Marlow, D. and Kodikara, J., 2009. Understanding
how the Australian Climate can Affect Pipe Failure. Proceedings of OzWater
09. AWA.
Gould, S. and Kodikara, J. K., 2008. Exploratory Statistical Analysis of Water

Reticulation Main Failures (Melbourne, Australia). Technical Report RR11.
Goulter, I. and Kazemi, A., 1988. Spatial and temporal groupings of water
main pipe breakage in Winnipeg. Canadian Journal of Civil Engineering,
15(1): 91–97.
Goulter, I. and Kazemi, A., 1989. Analysis of Water Distribution Pipe Failure
Types in Winnipeg, Canada. Journal of Transportation Engineering, 115(2):
95–111.
Grant, K., 1972. Terrain classification for engineering purposes of the Mel-
bourne area, Victoria, volume 11. CSIRO, Melbourne, Victoria, Australia.
Grant, K., 1973. Terrain classification for engineering purposes of the Queen-
scliff area, Victoria, volume 12. CSIRO, Melbourne, Victoria, Australia.
Grant, K. and Ferguson, T. G., 1978. Terrain analysis and classification for
engineering purposes of the Warragul area, Victoria, volume 21. CSIRO,
Melbourne, Victoria, Australia.
224
Groenevelt, P. H. and Bolt, G. H., 1972. Water Retention in Soil. Soil Science,
113(4): 238–245.
Groenevelt, P. H., Grant, C. and Semetsa, S., 2001. A new procedure to

determine soil water availability. Australian Journal of Soil Research, 39(3):
577–598.
Groenevelt, P. H. and Grant, C. D., 2001. Re-evaluation of the structural

properties of some British swelling soils. European Journal of Soil Science,
52(3): 469–477.
Groenevelt, P. H. and Grant, C. D., 2002. Curvature of shrinkage lines in

relation to the consistency and structure of a Norwegian clay soil. Geoderma,
106(3-4): 235–245.
Groenevelt, P. H. and Grant, C. D., 2004. Analysis of soil shrinkage data. Soil
and Tillage Research, 79(1): 71–77.
Habibian, A., 1994. Effect Of Temperature Changes On Water Main Breaks.

Journal of transportation engineering, 120(2): 10.
Holtz, W. and Gibbs, H., 1956. Engineering properties of exapnsive clays.

Transaction ASCE, 121: p641–677.
Hu, Y. and Hubble, D. W., 2007. Factors contributing to the failure of asbestos
cement water mains. Canadian Journal of Civil Engineering, 34(5): 608–
621.
Hudak, P. F., Sadler, B. and Hunter, B. A., 1998. Analyzing underground

water-pipe breaks in residual soils. Water Engineering and Management,
145(12): 5.
ICT, 2007. Soil Science Instrumentation.
Janssen, M., Zuidema, J. and Wanhill, R., 2004. Fracture Mechanics. Spon
Press, New York, 2nd edition.
Jarrett, R., Hussain, O. and van der Touw, J., 2002. Prediction of fail-
ures in water mains and prioritisation for replacements. Proceedings ofC-
SIRO/AWA 2nd Annual Seminar on Asset Management of Pipelines. 1–11.
Jarrett, R., Hussain, O. and van der Touw, J., 2003. Reliability assessment of
water pipelines using limited data,. Proceedings ofOzWater 03. Australian
Water Association.
Jarrett, R., Hussain, O., Veevers, A. and Van der Touw, J., 2001. Predic-
tive models for pipeline failures and identification of pipes for replacement.
Proceedings of Pipeline Asset Planning and Operations Workshop.
Kassiff, G. and Zeitlen, J. G., 1962. Behavior of pipes buried in expansive

clays. Journal of Soil Mechanics and Foundations division, Proceedings of
the American Society of Civil Engineers, 88(SM 2).
Kettler, A. and Goulter, I., 1985. An analysis of pipe breakage in urban water
distribution networks. Canadian Journal of Civil Engineering, 12: 286–293.
Kirkby, P., 1977. Internal Corrosion and Loss of Strength of Iron Pipes. Tech-
nical report, Water Research Centre.
Kleiner, Y. and Rajani, B., 2000. Considering time-dependent factors in the

statistical prediction of water main breaks. Proceedings of American Water
Works Association: Infrastructure Conference.
Kleiner, Y. and Rajani, B., 2001. Comprehensive review of structural deteri-

oration of water mains: statistical models. Urban Water, 3(3): 131–150.
Kleiner, Y. and Rajani, B., 2002. Forecasting Variations and Trends in Water-
Main Breaks. Journal of Infrastructure Systems, 8(4): 122–131.
Kodikara, J., Barbour, S. and Fredlund, D. G., 2002. Structure development

in surficial heavy clay soils: a synthesis of mechanisms. Australian Geome-
chanics: Journal and News of the Australian Geomechanics Society, 37(3):
25–40.
Kulhawy, F. and Mayne, P., 1990. Manual on estimating soil properties for
foundation design. Technical report, Electric Power Research Inst., Palo
Alto, CA (USA); Cornell Univ., Ithaca, NY (USA). Geotechnical Engineer-
ing Group.
Kuraoka, S., Rajani, B. and Zhan, C., 1996. Pipe soil interaction analysis
of field tests of buried PVC pipe. Journal of infrastructure systems, 2(4):
162–170.
Land Protection Service, 1985. Land Resources Data Atlas: Non Urban Areas.
Leong, E. C. and Rahardjo, H., 1997. Permeability Functions for Unsaturated

Soils. Journal of Geotechnical and Geoenvironmental Engineering, 123(12):
1118–1126.
226
Ma, Z. and Yamada, K., 1994. Durability evaluation of cast iron water sup-
ply pipes by sampling tests. Proceedings of Structural Engineering. Japan
Society of Civil Engineers.
Makar, J., 1999. Failure Analysis for Grey Cast Iron Water Pipes. Proceedings
of AWWA Distribution System Symposium.
Makar, J., Desnoyers, R. and McDonald, S., 2001. Failure modes and mech-
anisms in grey cast iron pipe. Proceedings of Underground infrastructure
research: Municipal, industrial and environmental applications.
Makar, J., Rogge, R., McDonald, S. and Tesfamariam, S., 2005. The Effect
of Corrosion Pitting on Circumferential Failures in Grey Cast Iron Pipes.
AWWARF, Denver, CO.
Makar, J. M. and McDonald, S. E., 2007. Mechanical behavior of spun-cast

gray iron pipe. Journal of Materials in Civil Engineering, 19(10): 826–833.
Makar, J. M., Rogge, R. and McDonald, S., 2002. Circumferential failures in

grey cast iron pipes. Proceedings of AWWA Infrastructure Conference.
Matti, M. A. and Al-Adeeb, A., 1985. Sulphate attack on asbestos cement

pipes. International Journal of Cement Composites and Lightweight Con-
crete, 7(3): 169–176.
McKenna, F., Fenves, G. L., Jeremic, B. and Scott, M. H., 2000. Open System
for Earthquake Engineering Simulation. http://opensees.berkeley.edu.
Moser, A. P., 2001. Buried Pipe Design. McGraw-Hill, 2 edition.
National Water Commission and WSAA, 2010. National Performance Report

2008-2009 Urban Water Utilities. Technical report.
Newport, R., 1981. Factors influencing the occurrence of bursts in iron water
mains. Water Supply and Management, 3: 274–278.
Olliff, J. and Rolfe, S., 2002. Condition assessment: the essential basis for best
rehabilitation practice. Proceedings of No-Dig 2002.
Olsen, P. A. and Haugen, L. E., 1998. A new model of the shrinkage charac-
teristic applied to some Norwegian soils. Geoderma, 83(1-2): 67–81.
Pelletier, G., Mailhot, A. and Villeneuve, J.-P., 2003. Modeling Water Pipe
Breaks—Three Case Studies. Journal of Water Resources Planning and
Management, 129(2): 115–123.
Peng, X. and Horn, R., 2005. Modeling soil shrinkage curve across a wide range
of soil types. Soil Science Society of America Journal, 69(3): 584–592.
Pham, H. and Fredlund, D., 2008. Equations for the entire soil-water charac-
teristic curve of a volume change soil. Canadian geotechnical journal, 45(4):
443–453.
Philadelphia Water Department, 1985. Water main pipe sample physical test-
ing program. Technical report.
Rajani, B. and Kleiner, Y., 2001. Comprehensive review of structural de-

terioration of water mains: physically based models. Urban Water, 3(3):
151–164.
Rajani, B., Makar, J., McDonald, S., Zhan, C., Kuraoka, S., Jen, C. and
Viens, M., 2000. Investigation of Grey Cast Iron Water Mains to Develop a
Methodology for Estimating Service Life. AWWARF, Denver, CO.
Rajani, B. and Tesfamariam, S., 2004. Uncoupled axial, flexural, and circum-
ferential pipe-soil interaction analyses of partially supported jointed water
mains. Canadian Geotechnical Journal, 41(6): 997.
Rajani, B. and Zhan, C., 1996. On the estimation of frost load. Canadian
Geotechnical Journal, 33(4): 629–641.
Rajani, B., Zhan, C. and Kuraoka, S., 1996. Pipe-soil interaction analysis of
jointed water mains. Canadian Geotechnical Journal, 33(3): 393–404.
Rajeev, P., Chan, D. and Kodikara, J. K., 2011. Field measurement of soil
moisture content with neutron moisture probe. Technical Report RR15.
Reeve, M. J. and Hall, D. G. M., 1978. Shrinkage in Clayey Subsoils of Con-

trasting Structure. European Journal of Soil Science, 29(3): 315–323.
Reynaud, A., 2010. Corrosion of Cast Irons. Elsevier, Oxford, 1737–1788.
Rixon, C., 1973. Geological Map of victoria.

228
Sadiq, R., Rajani, B. and Kleiner, Y., 2004. Fuzzy-Based Method to Evaluate
Soil Corrosivity for Prediction of Water Main Deterioration. Journal of
Infrastructure Systems, 10(4): 149–156.
Scott, R., 1990. Water Main Renewal Study: Reticulation Water Mains 1857-
1990. Technical report, Melbourne Water.
Seed, H., Woodard, R. and Lundgren, R., 1962. Prediction of swelling poten-
tial for compacted clays. Journal of soil mechanics and foundation division
ASCE, 88: p.53–87.
Seica, M. V. and Packer, J. A., 2004. Mechanical Properties and Strength

of Aged Cast Iron Water Pipes. Journal of Materials in Civil Engineering,
16(1): 69–77.
Seica, M. V. and Packer, J. A., 2006. Simplified Numerical Method to Evaluate

the Mechanical Strength of Cast Iron Water Pipes. Journal of Infrastructure
Systems, 12(1): 60–67.
Shamir, U. and Howard, C., 1979. An analytical approach to scheduling pipe

replacement. Journal AWWA, 71(5): 248–258.
Sillers, W. S. and Fredlund, D. G., 2001. Statistical assessment of soil-water

characteristic curve models for geotechnical engineering. Canadian geotech-
nical journal, 38(6): 1297–1313.
Smith, W., 1976. Frost loadings on underground pipe. Journal AWWA, 68(12):
673–674.
Standards Australia, 1993. AS1726 Geotechnical site investigations.
Standards Australia, 1995a. AS1289.3.2.1 Determination of the plastic limit

of a soil.
Standards Australia, 1995b. AS1289.3.6.2 Determination of the particle size

distribution of a soil -Analysis by sieving in combination with hydrometer
analysis (subsidiary method).
Standards Australia, 1996. AS2870 Residential slabs and footings-

Construction-Commentary.
Standards Australia, 2001. AS1289.1.1 Preparation of disturbed soil samples

for testing AS 1289.1.1.
Standards Australia, 2003. AS1289.3.6.3 Determination of the particle size

distribution of a soil -Standard method of fine analysis using a hydrometer.
Standards Australia, 2005. AS1289.2.1.1 Determination of the moisture con-

tent of a soil Oven drying method.
Standards Australia, 2008. AS1289.3.4.1 Determination of linear shrinkage of

a soil.
Standards Australia, 2009a. AS1289.3.1.1 Determination of the liquid limit of

a soil Four point Casagrande method.
Standards Australia, 2009b. AS1289.3.6.1 Determination of the particle size

distribution of a soil -Standard method of analysis by sieving.
Steinbrenner, W., 1934. Tafeln zur setzungsberechnung. Die Strasse, 1: 121–

124.
Stewart, I., 2007. Personal communication on usage era of pipe materials.
Talsma, T., 1977. A note on the shrinkage behaviour of a clay paste under
various loads. Australian Journal of Soil Research, 15: 275–277.
Tarantino, A., Ridley, A. and Toll, D., 2008. Field Measurement of Suction,
Water Content, and Water Permeability. Geotechnical and Geological En-
gineering, 26(6): 751–782.
Thomas, P., Baker, J. and Zelazny, L., 2000. An Expansive Soil Index for
Predicting Shrink-Swell Potential. Soil Science Society of America Journal,
64(1): 268–274.
Trickey, S. A. and Moore, I. D., 2005. Numerical Study of Frost-Induced

Ring Fractures in Cast Iron Water Pipes. GeoSask2005, 58th Canadian
Geotechnical Conference.
Trickey, S. A. and Moore, I. D., 2007. Three-Dimensional Response of Buried

Pipes under Circular Surface Loading. Journal of Geotechnical and Geoen-
vironmental Engineering, 133: 219.
Tripathy, S., Rao, K. S. S. and Fredlund, D. G., 2002. Water content-void ratio
swell-shrink paths of compacted expansive soils. Canadian Geotechnical
Journal, 39(4): 938–959.
230
van Genuchten, M. T., 1980. A Closed-form Equation for Predicting the Hy-
draulic Conductivity of Unsaturated Soils. Soil Sci Soc Am J, 44(5): 892–
898.
Victoria Department of Sustainability and Environment, 2006. Melbourne

Atlas. State of Victoria Department of Sustainability and Environment.
Vloerbergh, I. and Blokker, E., 2007. Failure data analysis - a Dutch case study.
Proceedings of 2nd Leading edge conference on strategic asset management.
Vu, H., 2002. Uncoupled and coupled solutions of volume change problems in
expansive soils. Doctoral thesis.
Vu, H. Q. and Fredlund, D. G., 2004. The prediction of one, two and three
dimensional heave in expansive soils. Canadian Geotechnical Journal, 41(4):
713–737.
Wakool Shire Council, 2000. D11 Development design specification - Water

reticulation.
Walski, T. and Pelliccia, A., 1982. Economic analysis of water main breaks.
Journal AWWA, 74(3): 140–147.
Water Services Association of Australia, 2003. Common Failure Modes in

Pressurised Pipeline Systems. Technical report, Water Services association
of Australia.
Water Services Association of Australia, 2007. Pipe System Acronyms.

https://www.wsaa.asn.au/Publications/NationalCodes/Documents/Pipe
Acronyms Classifications Issue 5.pdf.
Watkins, R. and Moser, A. P., 1998. Soil and surface loads on buried pipes
including minimum cover requirements. ASME 1998, 360.
Webb, T. H. and Gould, B. W., 1978. Water hammer. New South Wales
University Press, Kensington, N.S.W.
Wheeler, S. J., Sharma, R. S. and Buisson, M. S. R., 2003. Coupling of hy-

draulic hysteresis and stress-strain behaviour in unsaturated soils. Geotech-
nique, 53(1): 41–54.
Wightwick, M., 2010. Improvement of the performance of water-sensitive ge-

omaterials using hydrophobic additives. Ph.D. thesis.
Yamamoto, K., Mizoguti, S., Yoshimitsu, K. and Kawasaki, J., 1983. Relation
Between Graphitic Corrosion and Strength-Degradation of Cast Iron Pipe.
Corrosion Engineering, 32(3): 157–162.
Zhan, C. and Rajani, B., 1997. On the estimation of frost load in a trench:
theory and experiment. Canadian Geotechnical Journal, 34(4): 568–579.
Zhou, Q., Shimada, J. and Sato, A., 2001a. Three-Dimensional Spatial and
Temporal Monitoring of Soil Water Content Using Electrical Resistivity To-
mography. Water Resources Research, 37(2): 273–285.
Zhou, S. L., McMahon, T. A. and Wang, Q. J., 2001b. Frequency analy-

sis of water consumption for metropolitan area of Melbourne. Journal of
Hydrology, 247(1-2): 72–84.
Appendix A
Pipe network failure analysis

data preparation
A.1 Introduction
This appendix details the data preparation undertaken prior to the exploratory
statistical analysis. This process was undertaken in several parts described
below. The first part identifies and removes unreliable records from asset
data, part two identifies and removes unreliable records from failure data, part
three groups similar attributes within records and part four merges soil type
data based soil shrink/swell potential.
A.2 Asset data cleaning

Cleaning of the asset data was conducted using the nine tests listed below. It
should be noted that asset records identified by these tests, with the exception
of repeated asset ID test, were marked as such but not removed until all testing
had been completed to give an accurate indication of the quantity of erroneous
data. Asset records which failed the repeated asset ID check were removed from
the data before completing any of the other tests.
• Repeated asset ID
• No construction date test
• No material type
• No or short length test
233
234
• No diameter test
• No soil type test
• Large diameter test
• Construction date in future test
• Unlikely material type test
A.2.1 Repeated asset ID test

In order to ensure the integrity of the asset data each asset needs to have a
unique identifier (asset ID). A unique asset ID is also important for reliably
linking failure events to specific assets. The field chosen from the CWW data
to be the asset ID was ”Key”, based on information from CWW. However,
as there were several fields with names featuring ”ID” in the title this was
confirmed by linking to the failure data. It was found that the ”Key” field
could be linked to the greatest number of failure records than any of the other
fields tested. The field chosen from the SEWL data to be the asset ID was
”UNIQUE ID”, based on information from SEWL. As with the data from
CWW, this selection was also tested and confirmed as the field that could be
linked to the greatest number of failure records. As stated above, the CWW
and SEWL asset data were merged prior exporting to MS Access. Prior to
merging, the datasets were checked to ensure that no CWW asset ID was also
used by SEWL, no repetition of this kind was found. Once merged, the data
were then checked for repeated asset IDs.
Altogether 665 repeated asset IDs were found in the data (CWW 665;
SEWL 0). These asset IDs related to 2174 records, of which 731 records had
the asset ID ”0”. These 2174 records were further analysed to see if any
could be consolidated into single records. This was performed by identifying
which repeated asset IDs were only associated with a single construction date,
diameter and material. It was found that 1365 records, accounting for 634
asset IDs, could potentially be consolidated.
These 1365 records were then scrutinised spatially to determine whether
the assets they represented could be merged into a single continuous asset.
A merger was deemed possible if the assets were directly connected, this was
determined within GIS. If the assets with the repeated asset IDs existed and
were not directly connected, the asset(s) between them were scrutinised to
see if the asset(s) shared the same construction date, diameter and material
APPENDIX A. ANALYSIS DATA PREPARATION 235
despite asset ID(s) being different. If it was found that a group of assets could
be merged into a single asset, this was done and the new asset was assigned
the asset ID of the asset(s) with the greatest length (greatest total continuous
length by original asset ID). The new asset was assigned its new length as the
sum of its constituent parts and by definition retained their construction date,
diameter and material remained unchanged. To ensure that any associated
failures were not lost in this process the failure database was queried and the
asset ID of any failure assigned to any of the merged assets updated with the
new asset ID as appropriate.
Of the 634 asset IDs to be consolidated, only seven could not be completely
merged using the procedure described above. In these situations the asset
or merged assets with the greatest continuous length were retained and the
remaining asset(s) discarded. A summary of the asset data retained for further
cleaning is given in Table A.4. This data was then subjected to further tests
as detailed in the following sections.
Table A.1: Asset Data Summary after Repeated Asset ID Test

Authority Number of Failures Earliest Failure Date Most Recent Failure Date
CWW 36,634 06/02/1979 08/05/2007
SEWL 20,174 09/07/1996 20/03/2007
Total 56,808 06/02/1979 08/05/2007
A.2.1.1 No construction date test
Asset records which do not have a construction date cannot be deemed as

reliable and could not be included in the investigation. 451 assets (CWW 335;
SEWL 116) were found without a construction date. These assets were marked
and not included in the investigation.
A.2.1.2 No material type test
Assets that do not have any material type information cannot be deemed as
reliable and could not be included in the investigation. Four assets (CWW 4;
SEWL 0) were found to have no material type information, and these assets
were marked and not included in the investigation. It should be noted that
after grouping of the material types (see Section A.4.1) this number increased
to 100 (CWW 98; SEWL 2).
236
A.2.1.3 No or short length test
Assets with recorded lengths that were very small were deemed to provide
unreliable information and could not be included in the investigation. No
length information was directly provided by the water authorities for a large
number of assets. However as the assets data was provided in GIS format
the asset lengths were able to be calculated via a GIS algorithm. It should
be noted that assets length obtained this way may not be the true length as
some asset which may in reality curve, such as those supplying a court could
potentially have been drawn in the GIS as a short length between end points,
see Figure A.1.
Actual Length
Calculated Length
Figure A.1: Potential difference between calculated and actual asset length
Once length values were associated with all assets (either from values pro-
vided by water authority or via the GIS algorithm) these lengths were checked
to see if their length was sufficient to warrant their inclusion in the investiga-
tion. It was found that 356 assets (CWW 243; SEWL 113) had lengths less
than or equal to 0.1 m. A cut-off length of 0.1 m was chosen as it was deemed
very unlikely for an in-service asset to have a length this small. Therefore,
these assets were marked and not included in the investigation.
A.2.1.4 No diameter test
Assets with no recorded diameter (or a recorded diameter of ”0”) were deemed
to provide unreliable information and were not included in the investigation.
A total of 16 assets (CWW 14; SEWL 2) were found to have diameters of ”0”.
These assets were marked and not included in the investigation.
A.2.1.5 No soil type test
Assets which could not be assigned a soil type were deemed to provide unre-
liable information, and were not included in the investigation. The soil type
associated with each asset was determined using a GIS intersect algorithm.
Application of the intersect algorithm indicated that 825 assets (CWW 84;
SEWL 741) could not be assigned a no soil type. These assets were marked
and not included in the investigation. The specific details of how soil types
were allocated to pipe assets are described in the Soil Data section below.
A.2.1.6 Large diameter test
As this investigation focuses on the failure of reticulation pipes, those assets not
classified as reticulation pipes need to be excluded. One of the simplest ways
to do this uses the diameter of the asset. For the purpose of this investigation,
any asset with a diameter greater than or equal to 450 mm was assumed to be a
truck main. Altogether, 3,953 assets (CWW 3; SEWL 3,950) were determined
to be truck mains based on this requirement. These assets were marked and
not included in the investigation.
A.2.1.7 Construction date in future test
Asset records without a construction date were considered to provide unreliable

information and were not included in the investigation. No assets were found
without a construction date.
A.2.1.8 Unlikely material type test
There were 170 assets (CWW 665; SEWL 0) where the material type cited
was not in use at the time of the recorded construction date. Table A.4 shows
the years in which assets of various material types were first introduced to
Australia (Beech, 2007, Davis et al., 2008a, Scott, 1990, Stewart, 2007). Assets
which conflicted with this table were marked and not included in the analysis
as it was assumed that either the material or the construction dates had been
recorded incorrectly.
238
Table A.2: Material Introduction/Cessation Dates
Number prior to start date

Material Date Started Use Date Finished Use
CWW SEWL Total
PVC-U 1968 N/A 8 102 110
PE 1980 N/A 5 49 54
AC 1928 1980 7 563 570
DI 1975 N/A 78 136 214
CI N/A 1980 272 749 1021
S 1930 N/A 207 307 514
A.3 Failure data cleaning

Cleaning of the failure data was conducted using the six tests listed below.
• Repeated failure ID test
• No failure date test
• Failure date in future test
• Matching test
• Matched to erroneous asset test
• Preconstruction failure date test
As stated above, the failure data for CWW and SEWL were processed
separately before merging, as such the tests described below were applied to
the datasets independently. It should be noted that failure records identified
by these tests, with the exception of repeated failure ID test, were marked as
such but not removed until all testing had been completed to give an accurate
indication of the quantity of erroneous data. Failure records that failed the
repeated failure ID check were removed from the data before completing any
of the other tests.
A.3.1 Repeated failure ID test

In order to ensure the integrity of the failure data each failure needs to have
a unique identifier (failure ID). As with the asset data, repeated failure IDs
were not allowed as this indicates potential errors in the data recording; these
failures could not be retained for use in the analysis. Altogether 525 repeated
failure IDs (CWW 525; SEWL 0) were identified, accounting for 2642 of the
56808 failure records. These 2642 records were further analysed to ascertain if
any could be consolidated into single records. This was performed by identify-
ing which repeated failure IDs were only associated with a single failure date,
failure type and asset ID.
It was found that 1481 records, accounting for 299 failure IDs, could be
consolidated. Those failure records which could not be consolidated were dis-
carded. The remaining data were then subjected to further tests as detailed
in the following sections. A summary of the failure data retained for further
cleaning is given in Table A.4.
Table A.3: Failure Data Summary after Repeated Failure ID Test

Earliest Recorded Most Recent Recorded
Authority Number of Failures
Failure Date Failure Date
CWW 34,291 06/02/1979 08/05/2007
SEWL 20,174 09/07/1996 20/03/2007
Total 54,465 06/02/1979 08/05/2007
A.3.2 No failure date test

Failure records which do not have a failure date were considered to provide
unreliable information and were not included in the investigation. Four assets
(CWW 4; SEWL 0) were found without any failure date. These assets were
marked and not included in the investigation.
A.3.3 Failure date in future test

Failure records with a failure date in the future were considered to provide
unreliable information and could not be included in the investigation. No
failures had failure dates in the future.
A.3.4 Matching test

Failure records which could not be matched to an asset cannot be deemed as
reliable and could not be included in the investigation. 9,884 failures (CWW
5,334; SEWL 4,550) could not be matched to assets. These failures were
marked and not included in the investigation.
240
A.3.4.1 Matched to erroneous asset test
Failures that could be matched to an asset were tested to determine if the asset
they were matched with had failed any of the tests as described above. If it
was found that failures were matched to an unreliable asset, then these failures
were deemed as unreliable and could not be included in the investigation. 646
failures (CWW 263; SEWL 383) were matched to unreliable assets. These
failures were marked and not included in the investigation.
A.3.5 Preconstruction failure date test

The failures that could be matched to an asset were further tested to determine
if the asset they were matched with had a construction date after the recorded
failure date. A failure recorded against an asset before its construction could
indicate, that the failure was recorded against the wrong asset, that the failure
date is incorrect or that the failure was matched to an asset which has since
been replaced and the new asset assigned the same asset ID. Regardless of
the reason for the error in recording, these failures can also not be deemed as
reliable and could not be included in the investigation. 300 failures (CWW 178;
SEWL 122) were matched to assets which were constructed after the recorded
failure date. These failures were marked and not included in the investigation.
A.3.6 Merging of CWW and SEWL data

Prior to merging the failure data from CWW and SEWL, the failure IDs from
CWW and SEWL were compared. It was found that two datasets shared 1435
failure IDs. In order to maintain data integrity after merging the CWW failure
IDs were altered by adding 100,000,000 and the SEWL failure IDs were altered
by adding 900,000,000. These alterations ensured that the unique nature of
the failure IDs was maintained after the merger.
A.4 Data grouping

Examination of the data found that several of the input parameters to be used
during the investigation had a large number of values and that each value was
associated with only a small number of records. An example of this is failure
type from the CWW failure data. Often these values could be grouped to
provide a smaller number of values associated with a larger number of records.
It was found that there were 766 different descriptions of failure in this data.
However, examination of these descriptions found that a significant number

of the differing descriptions were the result of spelling errors and/or different
descriptions for the same failure type, such as ”CIRCUM FAIL”, ”BROKEN
BACK”, ”B/BACK” and ”CIRTCUM FAIL”. It is interesting to note that of
the 766 failure type descriptions in the failure data 502 of these descriptions
only appeared once.
A.4.1 Grouping of materials

The material data for CWW and SEWL contained 48 (CWW 26; SEWL 34)
different values for material. As stated above, it is possible to associate several
different values with a single value. The allocation of the provided material
types to grouped values was based on the WSAA classification system (Water
Services Association of Australia, 2007). A summary of the pipe materials as
given in the classification system are shown in Table A.4. Materials without
a known material type were mapped to a null value (i.e. left blank). The full
list of allocations is given in Gould and Kodikara (2008).
Table A.4: WSAA pipeline Acronyms (Water Services Association of Aus-

tralia, 2007)
Abbreviation Type
AC Asbestos Cement
CI Cast Iron
DI Ductile Iron
PE Polyethylene
S Steel
SA Steel above ground
GRP Glass Reinforced plastic
PE-X Cross-linked Polyethylene
PVC-M Modified Polyvinylchloride
PVC-O Oriented Polyvinylchloride
PVC-U Unplasticised Polyvinylchloride
ABS Acrylonitrile Butadiene Styrene
RC Reinforced Concrete
CU Copper
GWI Galvanised Wrought Iron
VC Vitrified Clay
A summary of the new material groups is shown in Table A.5. As an addi-

tional 96 assets were now allocated to the null material group, the No Material
Type Test was repeated and these assets were marked and not included in the
investigation.
242
Table A.5: Material group summary

Material Group Number of Assets Length of Assets (km)
100 1
AC 26,999 2,095
CI 66,568 3,958
CU 5,978 183
DI 31,525 1,601
GRP 10 2
GWI 1,418 85
PE 20,804 779
PVC-M 6,916 328
PVC-O 47 1
PVC-U 48,184 2,926
RC 594 53
S 12,503 628
A.4.2 Grouping of diameters

The diameter data for CWW and SEWL contained 50 (CWW 26; SEWL 48)
different values for diameter. The removal of the large diameters (≥ 450 mm)
and diameters of zero reduces this to 28 different values. Table A.6 shows
a summary of the number and length of assets for each diameter. It can be
calculated that there are nine diameters (shown in bold) accounting for >
98.5% of assets and > 99% of the asset length.
Table A.6: Diameter summary

Number of Length Number of Length
Diameter Diameter
assets (km) assets (km)
12 2 0.01 150 53,294 2,934.11
20 211 3.48 160 6 0.29
25 540 11.20 175 319 15.28
30 2 0.07 180 248 14.10
32 544 13.29 200 641 15.17
40 5,703 187.84 225 16,790 942.07
50 4,501 115.14 250 546 42.86
63 12,763 469.36 280 12 0.62
65 4 0.09 300 10,432 689.74
75 62 3.09 350 2 0.05
80 3,048 174.75 355 1 0.04
100 103,633 6,308.33 375 1,895 171.23
110 7 0.51 400 11 0.70
125 2,457 130.72
To simplify the analysis the diameters were grouped to reduce the number
of different diameters to a manageable number. Diameters were grouped so

that similar diameters remained together whilst obtaining as even as possible
spread in the length of asset in each group. Table A.7 shows how diameters
were grouped and a summary of the number and length of assets in each group.
Table A.7: Summary of diameters after grouping

Diameter Percentage
Diameter range Number of assets Length (km)
group (by length)
0-50 >0 → <50 7,002 215.90 1.8%
50-100 50 → <100 20,378 762.44 6.2%
100-150 100 → <150 106,100 6,439.86 52.6%
150-200 150 → <200 53,867 2,963.78 24.2%
200-250 200 → <250 17,431 957.24 7.8%
250-300 250 → <300 558 43.49 0.4%
300-350 300 → <350 10,432 689.74 5.6%
350-400 350 → ≤400 1,909 172.03 1.4%
A.4.3 Grouping of failure types

The failure type data for CWW and SEWL contained 849 (CWW 781; SEWL
100) different values for failure type. As stated above, it was seen that many
of these different values actually related to the same failure type. Table A.8
shows the three failure types to which the failures were allocated. The full
list showing how these allocations were made is given in Gould and Kodikara
(2008).
Table A.8: Summary of failure types after grouping

Failure Type Abbr. Failure Type Number of Failures
CF Circumferential Failure 13,883
LF Longitudinal Failure 6,851
OTHER Other 33,731
A.5 Soil data

Soil data were supplied by CWW and SEWL in GIS format. The specific
soil data provided were in the form of Grant engineering classification codes
(Grant codes) (Grant, 1972, 1973, Grant and Ferguson, 1978) and covered the
majority of the authorities’ supply areas. The two GIS maps were merged into
a single map, which covered a total area of 4,602.22 km2 .
244
The merged data contained 62 different Grant codes. The Grant codes clas-
sify soils using several different criteria; however, the focus of this investigation
is on the effect of soil shrink/swell potential on pipe failure. For this reason
the Grant code were reclassified to focus solely on soil shrink/swell potential.
Table A.9 shows the method used to reclassify the Grant codes to a simpler
and more relevant system.
A detailed description of the soil reclassifications is shown in Gould and
Kodikara (2008). After reclassification using the method detailed in Table A.9,
the new classifications were double-checked using the Land Resource Atlas -
Non Urban Areas (Land Protection Service, 1985). This reference was only
used as a check as it focused on non-urban areas and so did not contain in-
formation for the majority of the area of interest. After double-checking 12
reclassifications were updated, these changes can also be found in Grant Code
Reclassification. Figure A.2 shows the distribution of the different soil types
after final reclassification.
The 221,646 assets were allocated a soil shrink/swell potential using a GIS
spatial query. In situations where an asset existed in more than one soil the
asset was allocated the soil type that contained the majority of its length.
Assets which did not exist in a soil of known shrink/swell potential or in a soil
which had not been allocated a shrink/swell potential were given a null value,
i.e., left blank. Table A.10 shows the number and length of assets in each soil
type.
Table A.9: Soil reclassification scheme for grant codes

Dominant Grant
Shrink/swell
Code general Soil code Comments
potential
soil description
Very high
Description indicates that
Expansive soil shrink/swell VE
the soil is very expansive
potential
Mottled clay indicates that
the soil is has undergone
cycles of saturation and
High drying out (the resulting in
Mottled clay shrink/swell ME oxidation and colour
potential changes). This soil is
expansive as it will shrink
upon drying and swell
upon wetting
Clays other than those
Moderate
above are still generally
Clay shrink/swell EX
expansive but are likely to
potential
be more stable.
Silty soils are likely to be
Low
more expansive than sand
Silty soil shrink/swell SE
but still mostly
potential
non-expansive.
Negligible
Sand shrink/swell ST Sand is not expansive.
potential
The shrink/swell potential
Unknown Unknown of this soil type is either
unknown or not applicable
Inland
Inland Water of this soil type is either
Water
Fill Fill of this soil type is either
246
Figure A.2: Final soil classifications
Table A.10: Soil shrink/swell potential summary

Number of
Reactivity Code Length of Assets (km)
Assets
VE 43,325 2,349
ME 25,140 1,594
EX 79,956 4,274
SE 52,424 3,083
ST 19,976 1,306
Null 825 35
Appendix B
Field data
This appendix contains a summary of the adjusted data collected from the
field instrumentation site. The adjusted data shown was that produced by
application of calibration curves to the raw data as appropriate. Where data
were unavailable or has been quarantined from use within the body of this
thesis, this is indicated on the figures as appropriate. The data shown here was
collected from the times and dates detailed in Table 4.3 until the 17th of March
2010. The water reticulation pipe which was the focus of the instrumentation
is refereed to in this appendix as the pipe.
No effort was made to analyse the collected data presented here. The
presented data is accompanied with appropriate descriptions to identify the
type and source of the data presented.
B.1 Earth pressure

Two earth pressure cells were installed in Pits 2 and 3 beneath the pipe. The
response of the two earth pressure cells is shown in Figure B.1, along with the
calculated overburden (based on soil density and installation depth).
247
248
Pressure (kPa)
Overburden pressure Earth pressure cell 1 (Pit 2A) Earth pressure cell 2 (Pit 3A)
Figure B.1: Response of earth pressure cells in Pits 2 and 3 part A
B.2 Pipe water pressure

The pipe water pressure was monitored with a single gauge installed in the
pipe tapping instrumentation pit. Figure B.2 shows the collected data. Figure
B.2(a) shows the pipe water pressure on a daily basis. Data between the
19th of December 2008 and the 20th of March 2009 inclusive is not available
due to equipment malfunction. Figure B.2(b) shows the pipe water pressure
as measured every ten minutes. The data shown has been segregated into
weekday and weekend days to show the differing cyclical patterns which occur
at these times.
APPENDIX B. FIELD DATA 249
(erroneous data quarantined)

Equipment malfunction
Internal pressure (kPa)
(a) Pipe water pressure between the 12th of January 2008 and the 17th of March 2010
(b) Average daily fluctuation of pipe water pressure recorded for weekdays and weekends
Figure B.2: Pipe water pressure

250
B.3 Soil temperature

Soil temperature was measured at 15 locations by thermocouples installed in
Pits 1, 2 and 3. Four thermocouples were installed in Pit 1 above and below
the pipe at depths of 0.3 m, 0.55 m, 1.0 m and 1.75 m. Four thermocouples
were installed in each Pits
The temperature of the soil was measured at different depths at four lo-
cations by the thermocouples installed at each pit. Figures B.3 to B.6 show
the soil temperature for each location plotted with the variation of air tem-
perature measured by the weather station and pipe water temperature by the
temperature gauge connected to the pipe.
The data clearly show that when the soil is closer to ground surface (at
300 mm depth), it is more affected by the air temperature than soil at greater
depths (550 mm to 1750 mm). The effect of air temperature on soil tempera-
ture is most significant at Pit 1, as Figure B.3 shows that the soil temperature
at 300 mm depth follows the daily fluctuation of air temperature more closely
than the other locations. This may be due to the thermocouple at 300 mm
depth in Pit 1 being directly under the concrete driveway, which is more re-
sponsive to change of air temperature than the nature strip, as the thermal
conductivity of soil is less than that of concrete.
Figure B.3: Comparison of daily temperature at Pit 1


252
Figure B.6: Comparison of daily temperature at Pit 3 away from pipe

B.4 Soil volumetric water content and suction

The volumetric water content and matric suction of the soil were measured at
different depths at four locations by the sensors installed at each pit. Figure
B.7 through to Figure B.14 show the volumetric water content and matric
suction with time for each location. Whilst data were logged at ten minute
intervals, the results in these figures show daily averages.
Figure B.7: Daily average soil volumetric water content in Pit 1

254

Figure B.10: Daily average soil volumetric water content in Pit 3 away from
pipe
Figure B.11: Daily average soil suction in Pit 1

256

Figure B.14: Daily average soil suction in Pit 3 away from pipe
258
B.5 Weather station

The weather station recoded rainfall, air temperature and humidity data. Fig-
ure B.15 shows the rainfall recorded by the weather station. Air temperature
data are shown in Figures B.3 to B.6 and is not repeated here.
station installation
Prior to weather
Figure B.15: Daily rainfall recorded by the weather station
B.6 Soil movement

Soil movement was measured using the rod extensometer. Figure B.16 shows
the results of ground displacement at each anchor. The results for each anchor
and head were calculated using the initial values as the zero point. Displace-
ment was measured over the distance between Anchor 4, which was located
in the bed rock, and each anchor or the reference head. The amount of soil
displacement decreases with depth, with the reference head (400 mm depth)
showing the greatest displacement.
Prior to installation of
rod extensometer
Figure B.16: Daily soil movement
B.7 Pipe strain

Three sets of biaxial strain gauges installed on the pipe, the detailed locations
and labelling of the strain gauges are shown in Figure 4.15. Each set of strain
gauges consists of four biaxial gauges, one on the top and the bottom and
two at the spring line on opposite sides. Each of the biaxial gauges has one
gauge oriented along the longitudinal axis of the pipe and on gauge oriented
perpendicular to the longitudinal gauge to measure hoop (or circumferential)
strain.
The data from the strain gauges was collected at 10 minute intervals, how-
ever the data presented here are average daily values. The data were initialised
using the average daily value collected after instrumentation. It is important
to note that strain 3L (noted as longitudinal strain gauge at the bottom in
Pit 1), strain 1C (hoop strain gauge at the pipe top in Pit 1), and strain 12C
(hoop strain gauge at the left spring line in Pit 3) were malfunctioning and
were ignored.
Figures B.17 to B.19 depict the responses of longitudinal strain gauges
with time, in Pit 1, Pit 2 and Pit 3 respectively. The sign convention used is
that tension is positive and compression is negative, following the traditions
of structural engineering.
260
Quarantined data
Figure B.17: Daily average longitudinal strain at Pit 1
Quarantined data

Quarantined data
Quarantined data
Figure B.20: Daily average circumferential strain at Pit 1

262
Quarantined data
Quarantined data

Appendix C
Soil shrinkage curve equation

development
C.1 Introduction
The ideal shape of the soil shrinkage curve can be seen in Figure C.1. It can be
seen that in this ideal case the curve begins as a horizontal line before chang-
ing to rise parallel to the saturation line (Sr = 100%) before again reverting
to a horizontal line. These areas are refereed to as the residual shrinkage,
proportional shrinkage and structural shrinkage zones respectively.
From observation of the curve it is clear that the derivative of the curve
would exhibit a change in value from zero to the gradient of the curve in
the proportional zone and back to zero, see Figure C.2. A function which
displays two horizontal asymptotes with a rapid change between them is that
of tan−1 (x). A combination of two tan−1 (x) curves would create a function
which exhibits the desired shape and allows the gradient of the proportional
zone to be specified and the location at which the gradient changes to also be
specified.
The water content at which the gradient changes at the start and end of
the proportional zone are defined as a and b respectively. The gradient of the
proportional zone is defined as m. Finally as the rate of change in non-ideal
cases is variable between soils a fitting parameter is needed to accommodate
this requirement, this parameter is defined as φ.
Following this the equation for the derivative of the shrinkage curve is
shown in Equation C.1.
de m m
= − tan−1 (φ(ω − a)) + tan−1 (φ(ω − b)) (C.1)
dω π π
263
264
2.0
Structural
Change
ɸ
1.5
Proportional m
Change
1.0 .G S
e
Residual ω
ɸ =
Change ,e
0 %
10
=
S
r
0.5
er
b a
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

ω
Figure C.1: Idealised version of shrinkage curve
Equation C.1 can now be integrated to obtain the equation for the shrinkage
curve, Equation C.2.

m −1 1 2
e(ω) = − φ(ω − a)tan (φ(ω − a)) − ln(1 + (φ(ω − a)) )
φ.π 2
(C.2)
m −1 1 2
+ φ(ω − b)tan (φ(ω − b)) − ln(1 + (φ(ω − b)) ) + C
φ.π 2
The condition used to calculate the integration constant is shown in Equa-

tion C.2.
e(0) = e0 (C.3)
This condition is solved as shown in Equation C.4.

APPENDIX C. SOIL SHRINKAGE CURVE EQUATION DEVELOPMENT
265
Derivative of Void Ratio, de/dω

0.8
0.6
0.4
0.2
0
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4 0.45 0.5
Gravimetric Water Content, ω
Figure C.2: Idealised derivative of the shrinkage curve

m −1 1 2
C =er + (−a.φ)tan (−a.φ) − ln(1 + (−a.φ) )
φ.π 2
(C.4)
m −1 1 2
− (−b.φ)tan (−b.φ) − ln(1 + (−b.φ) )
φ.π 2
The final equation for the shrinkage curve is then shown in Equation C.5.

m −1 1 2
e(ω) = − φ(ω − a)tan (φ(ω − a)) − ln(1 + (φ(ω − a)) )
φ.π 2

m −1 1 2
+ φ(ω − b)tan (φ(ω − b)) − ln(1 + (φ(ω − b)) )
φ.π 2
(C.5)
m −1 1 2
+ er + (−a.φ)tan (−a.φ) − ln(1 + (−a.φ) )
φ.π 2

m −1 1 2
− (−b.φ)tan (−b.φ) − ln(1 + (−b.φ) )
φ.π 2
Finally, when Equation C.5 is used to describe the shrinkage curve, limits
are placed on the possible values for ω. ω is limited to values greater than or
266
equal to zero and to values where the calculated void ratio is not below the
100% saturation line, Equation C.6.
e
0≤ω≤ (C.6)
Gs
Figure C.3 shows the curve produced by Equation C.5 for a set of default
values. Figure C.4 shows the derivative of this curve. Due to the repetitive
nature of the final equation it becomes possible to simplify the presentation to
that shown by Equations C.7 and C.8

m −1 1 2
f (x, y) = − φ(x − y)tan (φ(x − y)) − ln(1 + (φ(x − y)) ) (C.7)
φ.π 2
e(ω) = f (ω, a) − f (ω, b) − f (0, a) + f (0, b) + er (C.8)
2.5
1.5
Void Ratio, e
0.5
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Figure C.3: Modelled shrinkage curve, where e0 = 0.5, a = 0.6, b = 0.1,

m = 2.5, φ = 100 and Gs = 2.5
2.5
1.5
de /d ω
0.5
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7
Figure C.4: Modelled shrinkage curve derivative, where e0 = 0.5, a = 0.6,

b = 0.1, m = 2.5, φ = 100 and Gs = 2.5

Monash 80129 PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Monash 80129 PDF

Uploaded by

Copyright:

Available Formats

A Study of the Failure of

Buried Reticulation Pipes in

Thesis in Fulfilment of the Requirements for the Degree of

Scott John Francis Gould

Department of Civil Engineering

To my Wife Jane, my Family and my Friends.

“If we knew what it was we were doing,

Scott John Francis Gould

I would like to thank my primary supervisor Associate Professor Jayantha

The following is a list of publications resulting from the research undertaken

Refereed journal papers

Refereed conference papers

Non-refereed conference papers

Gould, S. and Kodikara, J. K., 2009. Exploratory Statistical Analysis of Gas

Gould, S. and Kodikara, J. K., 2008. Exploratory Statistical Analysis of

Water Reticulation Main Failures (Melbourne, Australia). Technical Report

• αer - fitting parameter • φ - fitting parameter

• αa - fitting parameter • φae - fitting parameter represent-

• αω - coefficient of hydric soil ex- • φr - fitting parameter represent-

• ∆σgb - applied flexural stress • ψ ∗ - log10 (ψ)

• γ - soil unit weight (kN/m3 ) • ψmin - minimum suction value

• σ3 - pipe stress (P a) • d - notch, crack or pit depth (m)

• σgb - pipe stress (P a) • D - soil depth above the pipe

• e - soil void ratio

• a5 - constant • F - tensile force (N)

• a6 - constant • F1 - calculation parameter

• a7 - constant • F2 - calculation parameter

• A0 - calculation parameter • Fb - geometric correction factor

• A1 - calculation parameter • Gs - soil specific gravity

• h - layer height (m)

• Cd - load coefficient • i3 - geometric correction factor

• cr - corrosion rate (mm/year) • If - depth factor

• Is - shape factor • Pi - internal pressure

• j - soil layer • R2 - coefficient of determination

• KI - applied stress intensity fac- • S1 initial slope of the SWCC

• m0 - calculation parameter • t1 - fitting parameter

List of publications xiii

List of Figures xxix

List of Tables xxxvii

2 Factors affecting buried pipe failure 7

2.4.1.2 Dead load . . . . . . . . . . . . . . . . . . . . . 11

3 Pipe network historical failure analysis 29

3.4.4 Pipe material - pipe diameter interaction . . . . . . . . . 37

4 Field instrumentation of an in-service pipe in a high shrink/swell

4.3.4.1 Calibration of matric suction sensors . . . . . . 75

5 Mechanical properties of Australian Cast Iron reticulation pipes105

5.6 Results and Discussion . . . . . . . . . . . . . . . . . . . . . . . 112

6 A novel technique for model development 127

7 A void ratio - water content - net stress model for environ-

7.5.1 Theoretical understanding of fitting parameters in the

8 Modelling pipe-soil interaction 185

9 Conclusions and recommendations 215

9.3.5 Asset management application . . . . . . . . . . . . . . . 218

A Pipe network failure analysis data preparation 233

B Field data 247

C Soil shrinkage curve equation development 263

2.1 Manufacturing flaw in wall of PVC pipe . . . . . . . . . . . . . 9

3.1 CWW and SEWL Assets and Supply Boundaries . . . . . . . . 31

3.21 Intra-year failure rate variation by Soil . . . . . . . . . . . . . . 60

4.28 Weather station after installation at the site . . . . . . . . . . . 97

5.1 Tensile specimen dimensions . . . . . . . . . . . . . . . . . . . . 110

6.1 Idealised SWCC of volume change soil with labelled features

6.5 Parameter sensitivity results (x-axis in log scale). Note: Except